Integrated.SocialIntegrated.Social
Free AI planning tool

Free AI Token & API Cost Calculator

Compare current text-token rates across 93 verified model rows from OpenAI, Anthropic, Google, xAI, Kimi, Meta Muse, DeepSeek, Z.AI, Tencent, Mistral, Alibaba, Cohere. Manus is shown separately as a credit-based agent product, not as a made-up token API rate.

93 verified model rows12 API providersCache, batch and long-context rules where documented

Prices as of . Verified: .

Last amended: .

Need a different starting point? Browse all free AI marketing tools.

Standard text rates as of

Browse 93 catalog rows in US dollars per million text tokens at the documented default input tier. Direct DeepSeek rows use conservative peak pricing; the optional off-peak control applies only its verified alternative schedule. These are not a quote or a per-call total. Optional settings change model-card estimates, not this default-rate table. An empty prompt shows $0 input. Verify the provider page and conditions before you spend.

USD per million tokens. Leave either maximum blank for no limit; a maximum includes matching prices.

Sorting and price limits use standard uncached input/output rates at the documented default tier and peak pricing, not calculator settings.

Showing 93 of 93 models.

ProviderModelInput /MCached input /MOutput /MContext windowCatalog notes and documented options
OpenAIGPT-6 AstraFrontier · FlagshipVerify provider price ↗$10.00$1.00$50.001,050,000 tokens
  • Baseline retained. Public Responses and Chat Completions text API. Maximum input 922,000; maximum output 128,000. Standard global rates apply through 272,000 input tokens; long-context rates apply to the full request above that threshold. Cache writes, not reads: USD 12.50/1M for default 30m minimum lifetime or 25/1M for 12h; long-context writes 25/50. Flex has the published Batch rates, with slower responses and occasional resource unavailability. Fast prices shown are short-context only; its long-context table is 40 input/150 output/4 cached per 1M. Fast has no EU data residency and no latency SLA for Astra. Ultrafast is a separate published mode, not Standard.
  • Supplier API identity: gpt-6-astra.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: cachedInputPerM prices cache hits only, not cache writes. Residency/FedRAMP premiums are not included in global rates; see sourceNotes.
  • Long-context rate above 272,000 input tokens: Input $20.00/M; cached input $2.00/M; output $75.00/M.
  • Documented batch rate: Input $5.00/M; cached input $0.50/M; output $25.00/M.
  • Documented long-context batch rate: Input $10.00/M; cached input $1.00/M; output $37.50/M.
  • Documented fast rate: Input $20.00/M; cached input $2.00/M; output $100.00/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-6.1 SolBalanced · Latest SolVerify provider price ↗$2.00$0.10$10.001,050,000 tokens
  • Requested exact identity confirmed, distinct from gpt-6-sol. Public Responses and Chat Completions; tool calling requires Responses. Maximum input 922,000 and output 128,000. Cache writes USD 2.50/1M for 30m or 5/1M for 12h; long-context writes 5/10. Long-context rates cover the full API request above 272,000 input tokens. Flex shares published Batch pricing. Fast long-context rates are 8 input/30 output/0.40 cached per 1M. Ultrafast is separately priced at 6x Standard, not a discounted Standard rate. US and EU residency supported; Fast supports EU. Subscription usage in Codex/ChatGPT Work is not this API price scope.
  • Supplier API identity: gpt-6.1-sol.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: The supplied 15% reseller discount is unverified and is not applied. Cache-hit pricing alone omits cache-write cost.
  • Long-context rate above 272,000 input tokens: Input $4.00/M; cached input $0.20/M; output $15.00/M.
  • Documented batch rate: Input $1.00/M; cached input $0.05/M; output $5.00/M.
  • Documented long-context batch rate: Input $2.00/M; cached input $0.10/M; output $7.50/M.
  • Documented fast rate: Input $4.00/M; cached input $0.20/M; output $20.00/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-6.1 TerraBalanced · Latest TerraVerify provider price ↗$1.00$0.05$5.001,050,000 tokens
  • Requested exact public API identity confirmed. Text generation and reasoning; Responses and Chat Completions supported; tool calling requires Responses. Maximum input 922,000/output 128,000. Cache writes USD 1.25/1M for 30m or 2.50/1M for 12h; long-context writes 2.50/5. The >272,000-input-token surcharge applies to the full request. Flex shares Batch rates; Fast long-context rates 4 input/15 output/0.20 cached per 1M. US and EU residency supported, including Fast EU processing.
  • Supplier API identity: gpt-6.1-terra.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: The supplied 30% reseller discount is unverified and is not applied. Cache writes must be priced separately from cache hits.
  • Long-context rate above 272,000 input tokens: Input $2.00/M; cached input $0.10/M; output $7.50/M.
  • Documented batch rate: Input $0.50/M; cached input $0.025/M; output $2.50/M.
  • Documented long-context batch rate: Input $1.00/M; cached input $0.05/M; output $3.75/M.
  • Documented fast rate: Input $2.00/M; cached input $0.10/M; output $10.00/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-6 LunaBudget · High volumeVerify provider price ↗$0.10$0.01$0.501,050,000 tokens
  • Requested identity and baseline retained. Public text API; Responses and Chat Completions supported. Maximum input 922,000/output 128,000. Current model page now explicitly documents >272K input-token long-context trigger, resolving the baseline's missing-threshold warning. Cache writes USD 0.125/1M for 30m or 0.25/1M for 12h; long-context writes 0.25/0.50. Long rates apply to the full request. Flex shares Batch pricing. Fast long-context input/output/cached 0.40/1.50/0.04 per 1M. EU residency supports Standard, Fast, Flex and Batch. Chat Completions function calling requires reasoning_effort none; use Responses for built-in tools.
  • Supplier API identity: gpt-6-luna.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: The supplied 15% reseller discount is unverified and is not applied. Cache-hit price does not include cache writes.
  • Long-context rate above 272,000 input tokens: Input $0.20/M; cached input $0.02/M; output $0.75/M.
  • Documented batch rate: Input $0.05/M; cached input $0.005/M; output $0.25/M.
  • Documented long-context batch rate: Input $0.10/M; cached input $0.01/M; output $0.375/M.
  • Documented fast rate: Input $0.20/M; cached input $0.02/M; output $1.00/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-6 SolBalanced · Previous SolVerify provider price ↗$2.00$0.20$10.001,050,000 tokens
  • Baseline retained as its own currently documented API identity; do not relabel as GPT-6.1 Sol. Model page directs users to newer 6.1 Sol. Cache writes USD 2.50/1M for 30m or 5/1M for 12h; long writes 5/10. Full-request long tier above 272,000 input tokens. Maximum input 922,000/output 128,000. Flex shares Batch rates. EU residency supports Standard/Flex/Batch but not Fast. Fast long rates input/output/cached 8/30/0.80 per 1M.
  • Supplier API identity: gpt-6-sol.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Long-context rate above 272,000 input tokens: Input $4.00/M; cached input $0.40/M; output $15.00/M.
  • Documented batch rate: Input $1.00/M; cached input $0.10/M; output $5.00/M.
  • Documented long-context batch rate: Input $2.00/M; cached input $0.20/M; output $7.50/M.
  • Documented fast rate: Input $4.00/M; cached input $0.40/M; output $20.00/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-5.6 SolFrontier · PromoVerify provider price ↗$4.00$0.40$20.001,050,000 tokens
  • Baseline model retained; exact API identity uses gpt-5.6-sol, not baseline UI slug gpt-5-6-sol. gpt-5.6 is a routing alias, not a separately priced row. Public general text/reasoning model with Responses and Chat Completions supported. Cache writes USD 5/1M for 30m or 10/1M for 12h; long writes 10/20. Full-request long rates above 272,000 input tokens. Flex shares Batch rates. Fast long input/output/cached 16/60/1.60 per 1M. Maximum input 922,000/output 128,000.
  • Supplier API identity: gpt-5.6-sol.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: Official promotional rates are already the listed Standard rates, available at least through November 21, 2026. No additional discount applied and no post-November-21 rate guessed; recheck then.
  • Long-context rate above 272,000 input tokens: Input $8.00/M; cached input $0.80/M; output $30.00/M.
  • Documented batch rate: Input $2.00/M; cached input $0.20/M; output $10.00/M.
  • Documented long-context batch rate: Input $4.00/M; cached input $0.40/M; output $15.00/M.
  • Documented fast rate: Input $8.00/M; cached input $0.80/M; output $40.00/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-5.6 TerraBalancedVerify provider price ↗$2.00$0.20$12.001,050,000 tokens
  • Active, separately priced general text/reasoning tier, not an alias of 6.1 Terra. Maximum input 922,000/output 128,000. Cache writes USD 2.50/1M for 30m or 5/1M for 12h; long writes 5/10. Long-context rates cover full request above 272,000 input tokens. Flex shares Batch pricing. Fast long input/output/cached 8/36/0.80 per 1M.
  • Supplier API identity: gpt-5.6-terra.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Long-context rate above 272,000 input tokens: Input $4.00/M; cached input $0.40/M; output $18.00/M.
  • Documented batch rate: Input $1.00/M; cached input $0.10/M; output $6.00/M.
  • Documented long-context batch rate: Input $2.00/M; cached input $0.20/M; output $9.00/M.
  • Documented fast rate: Input $4.00/M; cached input $0.40/M; output $24.00/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-5.6 LunaBudgetVerify provider price ↗$0.20$0.02$1.201,050,000 tokens
  • Active, separately priced previous Luna tier. Maximum input 922,000/output 128,000. Cache writes USD 0.25/1M for 30m or 0.50/1M for 12h; long writes 0.50/1. Long rates cover full request above 272,000 input tokens. Flex shares Batch rates. Fast long input/output/cached 0.80/3.60/0.08 per 1M.
  • Supplier API identity: gpt-5.6-luna.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Long-context rate above 272,000 input tokens: Input $0.40/M; cached input $0.04/M; output $1.80/M.
  • Documented batch rate: Input $0.10/M; cached input $0.01/M; output $0.60/M.
  • Documented long-context batch rate: Input $0.20/M; cached input $0.02/M; output $0.90/M.
  • Documented fast rate: Input $0.40/M; cached input $0.04/M; output $2.40/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-5.5FrontierVerify provider price ↗$5.00$0.50$30.001,050,000 tokens
  • Active older distinct text/reasoning family; stable API alias rather than additional dated snapshot row. Maximum output 128,000. Long tier applies for the full session above 272,000 input tokens for Standard/Batch/Flex. No additional cache-write charge; only implicit caching; prompt_cache_retention supports 24h only (typically around 30m, up to 24h). Fast short-context table is shown; its long-context cells are unpriced, so no Fast long rate inferred.
  • Supplier API identity: gpt-5.5.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Long-context rate above 272,000 input tokens: Input $10.00/M; cached input $1.00/M; output $45.00/M.
  • Documented batch rate: Input $2.50/M; cached input $0.25/M; output $15.00/M.
  • Documented long-context batch rate: Input $5.00/M; cached input $0.50/M; output $22.50/M.
  • Documented fast rate: Input $12.50/M; cached input $1.25/M; output $75.00/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-5.5 ProFrontier · ProVerify provider price ↗$30.00Not listed$180.001,050,000 tokens
  • Active, distinct higher-compute Pro tier. Responses and Batch supported; Chat Completions unsupported. Maximum output 128,000. No cached-input discount; cache rate omitted, not zero. No additional cache-write charge; 24h retention only. Explicit pricing-document context footnote supplies the >272,000-input threshold for Standard long rates. Batch long-context cells are '-' and not inferred.
  • Supplier API identity: gpt-5.5-pro.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Long-context rate above 272,000 input tokens: Input $60.00/M; cached input not listed; output $270.00/M.
  • Documented batch rate: Input $15.00/M; cached input not listed; output $90.00/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-5.4BalancedVerify provider price ↗$2.50$0.25$15.001,050,000 tokens
  • Active older general coding/reasoning family; maximum output 128,000. >272,000-input long tier applies to full session for Standard/Batch/Flex. No additional cache-write charge; supports older in_memory/24h retention. Batch short cached rate is published as USD 0.13/1M (table rounding preserved, not replaced by inferred 0.125). Fast long cells are unpriced; no Fast long rate inferred.
  • Supplier API identity: gpt-5.4.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Long-context rate above 272,000 input tokens: Input $5.00/M; cached input $0.50/M; output $22.50/M.
  • Documented batch rate: Input $1.25/M; cached input $0.13/M; output $7.50/M.
  • Documented long-context batch rate: Input $2.50/M; cached input $0.25/M; output $11.25/M.
  • Documented fast rate: Input $5.00/M; cached input $0.50/M; output $30.00/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-5.4 ProFrontierVerify provider price ↗$30.00Not listed$180.001,050,000 tokens
  • Active distinct Pro tier; Responses supported, Chat Completions unsupported. Maximum output 128,000. No cached-input rate published; omit rather than substitute zero. >272,000-input-token long tier applies to full session. Earlier models have no additional cache-write charge.
  • Supplier API identity: gpt-5.4-pro.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: Official pricing Markdown lists Batch rates, but the current model page marks Batch Not supported. Batch fields omitted because applicability conflicts; no inference call made to resolve it.
  • Long-context rate above 272,000 input tokens: Input $60.00/M; cached input not listed; output $270.00/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-5.4 MiniBudgetVerify provider price ↗$0.75$0.075$4.50400,000 tokens
  • Active separately priced mini coding/subagent model; maximum input 272,000/output 128,000. No long tier published for this model. Earlier-model implicit caching with no additional write charge; do not assume GPT-5.6 30m/12h write billing applies.
  • Supplier API identity: gpt-5.4-mini.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Documented batch rate: Input $0.375/M; cached input $0.0375/M; output $2.25/M.
  • Documented fast rate: Input $1.50/M; cached input $0.15/M; output $9.00/M.
  • Documented US inference multiplier: 1.1×.
OpenAIGPT-5.2BalancedVerify provider price ↗$1.75$0.175$14.00400,000 tokens
  • Active previous flagship, not tagged deprecated in current full catalog. Maximum output 128,000. No long-context surcharge published for this 400,000-context identity. Earlier implicit caching has no additional write charge; model is listed as supporting in_memory and extended 24h retention, subject to organization settings. Do not apply post-March-5-2026 model-release residency uplift automatically to this December-2025 family.
  • Supplier API identity: gpt-5.2.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Documented batch rate: Input $0.875/M; cached input $0.0875/M; output $7.00/M.
  • Documented fast rate: Input $3.50/M; cached input $0.35/M; output $28.00/M.
OpenAIGPT-5.2 ProFrontierVerify provider price ↗$21.00Not listed$168.00400,000 tokens
  • Active previous Pro tier; stable API alias, no duplicate dated snapshot. Responses supported, Chat Completions unsupported; maximum output 128,000. Cached input is '-' in pricing and omitted, not zero. No published long tier for this context size.
  • Supplier API identity: gpt-5.2-pro.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: Official pricing Markdown lists Batch rates, but current model endpoint support marks Batch Not supported. Batch fields omitted because applicability conflicts.
OpenAIChat LatestBalancedVerify provider price ↗$5.00$0.50$30.00400,000 tokens
  • Distinct priced public API identity for the latest ChatGPT Instant model. Responses and Chat Completions supported; Batch unsupported. Maximum input 272,000/output 128,000. No long-context tier published. Underlying snapshot changes regularly. Cache-write/TTL rules cannot safely be inferred from this moving alias's name; confirm applicable model behavior before estimating writes.
  • Supplier API identity: chat-latest.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: OpenAI does not recommend this moving ChatGPT model for production API use and instead recommends GPT-6 Astra. Numeric cached-input price alone does not establish cache-write cost or retention eligibility.
OpenAIGPT-4oLegacy · DeprecatedVerify provider price ↗$2.50$1.25$10.00128,000 tokens
  • Published baseline legacy retained because current official pricing and API model page still document it. Stable alias maps to gpt-4o-2024-08-06; no separate dated alias row. Responses/Chat Completions/Batch supported today. Maximum output 16,384. Earlier implicit cache has no extra write charge; in_memory entries generally 5–10 minutes inactive, up to one hour. No published Batch cached rate ('-'), so omitted rather than assumed zero.
  • Supplier API identity: gpt-4o.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: Deprecated, with API shutdown January 14, 2027 for currently served August/November 2024 snapshots. The older May-2024 snapshot shut down October 23, 2026 and is not represented here. Recheck/remove before scheduled shutdown.
  • Documented batch rate: Input $1.25/M; cached input not listed; output $5.00/M.
  • Documented fast rate: Input $4.25/M; cached input $2.125/M; output $17.00/M.
OpenAIGPT-4o miniLegacy · DeprecatedVerify provider price ↗$0.15$0.075$0.60128,000 tokens
  • Published baseline legacy retained with currently documented Standard rates. Stable alias maps to gpt-4o-mini-2024-07-18; maximum output 16,384. Responses/Chat Completions/Batch supported. Earlier implicit cache: no extra write charge; in_memory generally 5–10 minutes inactive, up to one hour. No published Batch cached rate ('-'), so omitted rather than assumed zero.
  • Supplier API identity: gpt-4o-mini.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: Deprecated; scheduled removal from API January 14, 2027. Recheck/remove before shutdown.
  • Documented batch rate: Input $0.075/M; cached input not listed; output $0.30/M.
  • Documented fast rate: Input $0.25/M; cached input $0.125/M; output $1.00/M.
AnthropicClaude Fable 5.1Frontier · FrontierVerify provider price ↗$10.00$0.25$50.001,000,000 tokens
  • Baseline identity, standard rates, batch rates and context revalidated. Public Claude API model; adaptive thinking is always on. Standard global/default synchronous text rates in USD per million tokens. cachedInputPerM prices cache hits/refreshes, not writes: 5-minute writes cost $12.50/MTok and 1-hour writes $20/MTok. Minimum cacheable prefix is 512 tokens. Full 1M context has no long-context premium. US-only inference is an optional 1.1x modifier on every token category, not the default. Batch cache rate follows the documented 50% discount on all usage; cache hits are best effort.
  • Supplier API identity: claude-fable-5-1.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:17Z.
  • Data-use warning: General public access is distinct from Claude Mythos 5.1 verification-only access. Safeguards can restrict some cybersecurity and biology workloads. Cache write/TTL costs must be accounted for separately.
  • Documented batch rate: Input $5.00/M; cached input $0.125/M; output $25.00/M.
  • Documented US inference multiplier: 1.1×.
AnthropicClaude Opus 5.5Frontier · FrontierVerify provider price ↗$4.00$0.20$20.001,000,000 tokens
  • Baseline identity and all token rates revalidated. Public Claude API model with always-on adaptive thinking. Standard global/default synchronous text rates in USD per million tokens. Cache hits are not cache writes: standard 5-minute writes $5/MTok; 1-hour writes $8/MTok; minimum cacheable prefix 512 tokens. Full 1M context has no long-context premium. Fast mode uses the same model with speed=fast, is now stable on Opus 5.5, and is not a separate model row. Fast cache hit $0.40/MTok follows the documented 0.05x multiplier on $8/MTok fast input; fast cache writes are $10/MTok for 5m and $16/MTok for 1h. US-only inference multiplies all token categories by 1.1. Batch cache hit rate is $0.10/MTok under the documented 50% all-usage discount.
  • Supplier API identity: claude-opus-5-5.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:17Z.
  • Data-use warning: Corrected the baseline note: Opus 5.5 fast mode is stable, not research preview. Fast mode cannot be combined with Batch API or Priority Tier commitments. Supplied reseller discount is not verified. Cache write/TTL costs remain separate.
  • Documented batch rate: Input $2.00/M; cached input $0.10/M; output $10.00/M.
  • Documented fast rate: Input $8.00/M; cached input $0.40/M; output $40.00/M.
  • Documented US inference multiplier: 1.1×.
AnthropicClaude Sonnet 5.5Balanced · BalancedVerify provider price ↗$2.00$0.10$10.001,000,000 tokens
  • Added the latest current Sonnet tier. Public Claude API model; standard global/default synchronous text rates in USD per million tokens. Cache read price was reduced to $0.10/MTok on October 7, 2026; do not use the earlier $0.20/MTok launch rate. Cache hits do not price cache writes: standard 5m writes $2.50/MTok and 1h writes $4/MTok; minimum cacheable prefix 512 tokens. Full 1M context is standard-priced. Optional US-only inference multiplies all token categories by 1.1. Batch cache hit is $0.05/MTok under the documented 50% all-usage discount.
  • Supplier API identity: claude-sonnet-5-5.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:17Z.
  • Documented batch rate: Input $1.00/M; cached input $0.05/M; output $5.00/M.
  • Documented US inference multiplier: 1.1×.
AnthropicClaude Haiku 5.5Budget · BudgetVerify provider price ↗$0.10$0.01$0.501,000,000 tokens
  • Added requested/latest Haiku model with exact public Claude API identity. Standard global/default synchronous text rates in USD per million tokens. Base rates apply to prompts up to and including 100,000 tokens. A request with more than 100,000 total prompt/input tokens uses the higher input, output, cache-read and cache-write rates for that entire request, not just excess tokens. Prompt length includes uncached input, cache reads and cache writes; output length is not the trigger. Each request is assessed independently, and earlier requests are not repriced. Up to 100k: 5m writes $0.125/MTok; 1h writes $0.20/MTok. Over 100k: 5m writes $0.625/MTok; 1h writes $1/MTok. Cacheable minimum prefix 512 tokens. Batch table documents both tiers; cache hits follow the documented 50% all-usage discount. US-only inference applies 1.1x to every category in both prompt-length tiers.
  • Supplier API identity: claude-haiku-5-5.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:17Z.
  • Data-use warning: The supplied $0.10/$0.50/$0.01 list rates are verified only for prompts of at most 100,000 tokens, not the full 1M context. Cache write costs are separate. Supplied reseller discount is not verified.
  • Long-context rate above 100,000 input tokens: Input $0.50/M; cached input $0.05/M; output $2.50/M.
  • Documented batch rate: Input $0.05/M; cached input $0.005/M; output $0.25/M.
  • Documented long-context batch rate: Input $0.25/M; cached input $0.025/M; output $1.25/M.
  • Documented US inference multiplier: 1.1×.
AnthropicClaude Opus 5Legacy · LegacyVerify provider price ↗$5.00$0.50$25.001,000,000 tokens
  • Preserved documented baseline legacy model: direct API active (legacy), with migration to Opus 5.5 recommended. Standard global/default text rates in USD per million tokens. 5m cache writes $6.25/MTok, 1h writes $10/MTok, minimum prefix 512 tokens. Full 1M context at standard pricing. Optional US-only inference 1.1x applies across categories. Fast mode remains research preview, requires fast-mode-2026-02-01 beta header and account-manager access or waitlist; not a broadly guaranteed default mode. Fast cache hit $1/MTok follows documented 0.1x of $10/MTok fast input; fast cache writes $12.50/MTok for 5m and $20/MTok for 1h. Batch cache hit is $0.25/MTok under the official 50% all-usage discount.
  • Supplier API identity: claude-opus-5.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:17Z.
  • Data-use warning: Retained only because it is a documented baseline legacy row. Fast mode has access conditions and cannot combine with Batch API or Priority Tier. Cache writes are not covered by the cache-hit rate.
  • Documented batch rate: Input $2.50/M; cached input $0.25/M; output $12.50/M.
  • Documented fast rate: Input $10.00/M; cached input $1.00/M; output $50.00/M.
  • Documented US inference multiplier: 1.1×.
AnthropicClaude Sonnet 5Legacy · LegacyVerify provider price ↗$2.00$0.20$10.001,000,000 tokens
  • Preserved documented baseline legacy model: direct API active (legacy), with migration to Sonnet 5.5 recommended. Standard global/default text rates in USD per million tokens. $2/$10 input/output prices are now permanent standard prices: the announced introductory period ended August 31, 2026, but the scheduled September 1 price increase was canceled. Cache writes remain $2.50/MTok for 5m and $4/MTok for 1h; cache-hit rate $0.20/MTok is distinct from Sonnet 5.5's $0.10. Minimum cacheable prefix 1,024 tokens. Full 1M context has no long-context premium. US-only inference is optional 1.1x across all token categories. Official batch cache hit is $0.10/MTok.
  • Supplier API identity: claude-sonnet-5.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:17Z.
  • Documented batch rate: Input $1.00/M; cached input $0.10/M; output $5.00/M.
  • Documented US inference multiplier: 1.1×.
AnthropicClaude Haiku 4.5Legacy · LegacyVerify provider price ↗$1.00$0.10$5.00200,000 tokens
  • Preserved baseline legacy model with corrected stable API snapshot: baseline claude-haiku-4-5 remains a documented convenience alias resolving to claude-haiku-4-5-20251001; only the pinned identity is listed. Official context is 200,000 tokens, not baseline 1M. Direct API status is active (legacy), still available despite the earliest retirement commitment date having passed. Standard global/default text rates in USD per million tokens; standard and batch rates revalidated. Cache writes $1.25/MTok for 5m and $2/MTok for 1h, minimum cacheable prefix 4,096 tokens. Batch cache hits $0.05/MTok follow the documented 50% all-usage discount. No first-party US-residency multiplier: this model predates Claude 4.6 and does not support inference_geo.
  • Supplier API identity: claude-haiku-4-5-20251001.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:17Z.
  • Data-use warning: Corrected baseline contextTokens from 1000000 to 200000 and removed unsupported usResidencyMultiplier=1.1. Sending inference_geo to Haiku 4.5 returns a 400 error. Cache-hit cost alone does not price cache creation.
  • Documented batch rate: Input $0.50/M; cached input $0.05/M; output $2.50/M.
GoogleGemini 4 ArgonFrontier · Intro price through 2026-11-30Verify provider price ↗$2.00$0.10$10.001,048,576 tokens
  • Exact stable API identity: gemini-4-argon. Generally available since October 13, 2026; paid Gemini Developer API, not an invite-only or managed-agent price. Standard introductory USD/1M input/output/cache-hit rates apply through November 30, 2026; published December 1 rates are $4/$20/$0.20. Published Batch rates thereafter are $2/$10/$0.10. The 1,048,576-token context window is unified: input, thinking, and output share it. Synchronous output defaults to 65,536 tokens; long decode can use the remaining unified window. No prompt-length price tier is documented.
  • Supplier API identity: gemini-4-argon.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: Implicit caching only; explicit cache writes/storage are unsupported. Minimum implicit-cache input is 4,096 tokens; hits are not guaranteed. In multi-turn/agent loops, the full previous request is cached only when the new request adds at least 2,050 tokens; otherwise its last 2,050 tokens are reprocessed at the standard input rate. Output pricing includes thinking tokens. All processed tokens, including internal grounding tool instructions and function declarations, are billed. Grounding/tool charges are additional. No Free-tier access.
  • Documented batch rate: Input $1.00/M; cached input $0.05/M; output $5.00/M.
GoogleGemini 3.8 FlashBalanced · Promo through 2026-12-31Verify provider price ↗$0.75$0.075$3.751,048,576 tokens
  • Exact stable API identity: gemini-3.8-flash; baseline calculator ID preserved. Current default paid Standard rates apply through December 31, 2026. Google publishes January 1, 2027 Standard input/output/cache-hit rates of $1.50/$7.50/$0.15 per 1M and Batch rates $0.75/$3.75/$0.075. Explicit-cache storage is an additional $0.50 per 1M cached tokens per hour through December 31, then $1.00 per 1M per hour, in both Standard and Batch tables. No separate prompt-length pricing tier is documented.
  • Supplier API identity: gemini-3.8-flash.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: Cache-hit rate is not an all-in cache-write/storage price. Implicit caching requires at least 4,096 input tokens and does not guarantee hits. Explicit caching is Beta and available through GenerateContent, not Interactions; default TTL is one hour, with token-size and TTL-dependent storage charges. Developer documentation does not separately specify an explicit cache-creation/write token rate; do not assume it is free or import Cloud-specific billing. Output includes thinking; tools/grounding are extra.
  • Documented batch rate: Input $0.375/M; cached input $0.0375/M; output $1.875/M.
GoogleGemini 3.6 FlashLegacy · Previous generation; promo through 2026-12-31Verify provider price ↗$0.75$0.075$3.751,048,576 tokens
  • Exact stable API identity: gemini-3.6-flash; baseline calculator ID preserved. Google explicitly says this previous-generation model remains available for production use, and no shutdown is announced. Current Standard rates apply through December 31, 2026; January 1, 2027 Standard input/output/cache-hit rates are $1.50/$7.50/$0.15 per 1M, Batch $0.75/$3.75/$0.075. Standard and Batch explicit-cache storage is additional $0.50 per 1M tokens per hour through December 31, then $1.00. No prompt-length price tier is documented.
  • Supplier API identity: gemini-3.6-flash.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: Cache-hit rate excludes explicit storage and does not establish cache-write pricing. Implicit-cache minimum 4,096 tokens; hits not guaranteed. Explicit caching is Beta on GenerateContent only; default TTL one hour, storage billed by cached-token count and TTL. No distinct explicit-write rate established by the developer docs inspected. Output includes thinking; tool/grounding charges are additional.
  • Documented batch rate: Input $0.375/M; cached input $0.0375/M; output $1.875/M.
GoogleGemini 3.1 Pro PreviewFrontier · Preview; long-context tierVerify provider price ↗$2.00$0.20$12.001,048,576 tokens
  • Exact publicly documented preview API identity gemini-3.1-pro-preview; no invented stable gemini-3.1-pro alias. Baseline calculator ID preserved. Base prices apply when prompt size is <=200,000 tokens; strictly >200,000 prompt tokens selects the higher input, output, and cache-hit prices for that request. Batch cache-hit rates are explicitly the same as Standard, not automatically discounted. Explicit-cache storage adds $4.50 per 1M cached tokens per hour in both context tiers and both Standard and Batch. The customtools endpoint is documented at the same prices and is not duplicated as a distinct priced tier.
  • Supplier API identity: gemini-3.1-pro-preview.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: Preview, not GA; no shutdown date announced. No Free-tier API access. Output prices include thinking tokens. Implicit-cache minimum 4,096 tokens, hits not guaranteed. Explicit caching is Beta through GenerateContent only; default TTL one hour. Cache-hit numbers exclude storage and do not establish an explicit cache-write rate. Cached tokens count in prompt/context limits; non-cached inputs and tools are separately charged.
  • Long-context rate above 200,000 input tokens: Input $4.00/M; cached input $0.40/M; output $18.00/M.
  • Documented batch rate: Input $1.00/M; cached input $0.20/M; output $6.00/M.
  • Documented long-context batch rate: Input $2.00/M; cached input $0.40/M; output $9.00/M.
GoogleGemini 3.5 Flash-LiteBudget · BudgetVerify provider price ↗$0.30$0.03$2.501,048,576 tokens
  • Exact stable API identity gemini-3.5-flash-lite; baseline calculator ID preserved. Text-scope Standard rate is $0.30 input, $2.50 output including thinking, $0.03 cache-hit per 1M tokens. Explicit-cache storage adds $1.00 per 1M tokens per hour for Standard and Batch. No prompt-length tier or promotion expiry is documented. Batch cached-input rate corrected from baseline $0.02 to $0.015 using repeated fresh GETs of the official migrated pricing Markdown.
  • Supplier API identity: gemini-3.5-flash-lite.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: Source-representation discrepancy: the initial ai.google.dev extraction and an earlier GET of aistudio pricing showed Batch cache $0.02, but fresh GETs of aistudio.google.com/docs/pricing.md at 22:20:22 and 22:20:36 GMT showed $0.015. Current raw Markdown value used; not an inferred 50% discount. Cache-hit rate excludes explicit storage and does not price cache writes. Implicit caching is enabled for Gemini 2.5 and newer, but this model's exact minimum is absent from the inspected developer cache table; no minimum guessed. Explicit caching Beta on GenerateContent; default TTL one hour. Tools/grounding are additional.
  • Documented batch rate: Input $0.15/M; cached input $0.015/M; output $1.25/M.
GoogleGemini 3.1 Flash-LiteBudget · BudgetVerify provider price ↗$0.25$0.025$1.501,048,576 tokens
  • Exact stable API identity gemini-3.1-flash-lite; baseline calculator ID preserved. These are text input and text output prices, not the separate audio input prices. Standard explicit-cache storage adds $1.00 per 1M tokens per hour; Batch table explicitly lists $0.50 per 1M tokens per hour. No prompt-length pricing tier or promo expiry documented. Earliest published shutdown May 7, 2027; recommended replacement gemini-3.5-flash-lite.
  • Supplier API identity: gemini-3.1-flash-lite.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: Output includes thinking. Cache-hit rate excludes storage and does not establish explicit cache-write cost. Implicit caching applies to Gemini 2.5 and newer but exact model minimum is not in the inspected developer cache table; not guessed. Explicit caching Beta on GenerateContent only, default TTL one hour. Additional non-cached tokens, tools, and grounding charges are excluded from listed token rates.
  • Documented batch rate: Input $0.125/M; cached input $0.0125/M; output $0.75/M.
GoogleGemini 2.5 Flash-LiteLegacy · Legacy; existing users onlyVerify provider price ↗$0.10$0.01$0.401,048,576 tokens
  • Exact stable API identity gemini-2.5-flash-lite; documented baseline legacy retained, not added as a newest/general-availability recommendation. Google limits 2.5 access to users who actively used it in the past. This model is explicitly not deprecated and continues through the API until further notice. Standard and Batch explicit storage add $1.00 per 1M cached tokens per hour; Batch cache-hit rate remains $0.01, not automatically halved. Prices are text scope, excluding separate audio input rates. No prompt-length price tier is documented.
  • Supplier API identity: gemini-2.5-flash-lite.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02.
  • Data-use warning: Not generally available to new API users/projects; cannot be used as an unrestricted current-model recommendation. Cache-hit price excludes storage and does not establish explicit creation/write pricing. Exact model minimum is absent from inspected developer cache table; not guessed. Explicit caching Beta through GenerateContent only; default TTL one hour. Output includes thinking, and tools/grounding cost extra.
  • Documented batch rate: Input $0.05/M; cached input $0.01/M; output $0.20/M.
xAIGrok 4.7Frontier · FrontierVerify provider price ↗$2.00$0.50$6.00500,000 tokens
  • Exact public API model: grok-4.7. Global/default standard non-batch text rates. Latest flagship for coding and general work; configurable reasoning. All tokens in a request use long-context rates once total prompt tokens, including cached tokens, reach the threshold. Batch is supported with supplier-confirmed 20% discount. US endpoint is supported; the 1.1 multiplier is separate from the global base rates. Cache rates cover hits only; see shared cache and retention sourceNotes.
  • Supplier API identity: grok-4.7.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:34Z.
  • Data-use warning: The authoritative pricing table and its fresh Markdown explicitly use >=200,000 prompt tokens, encoded here as inclusive. Model-page prose and cache usage prose instead say 'exceed'; this documentation inconsistency remains flagged. Grok 4.7 Fast is not a public API model and is excluded.
  • Long-context rate at or above 200,000 input tokens: Input $4.00/M; cached input $1.00/M; output $12.00/M.
  • Documented batch rate: Input $1.60/M; cached input $0.40/M; output $4.80/M.
  • Documented long-context batch rate: Input $3.20/M; cached input $0.80/M; output $9.60/M.
  • Documented US inference multiplier: 1.1×.
xAIGrok 4.6Frontier · FrontierVerify provider price ↗$2.00$0.50$6.00500,000 tokens
  • Exact public API model: grok-4.6. Still documented and priced, so the baseline row is retained. Global/default text rates; batch supported with official 20% discount. US endpoint supported with separate 1.1 token premium. Reasoning is charged at output rate. Shared cache and region conditions are in sourceNotes.
  • Supplier API identity: grok-4.6.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:34Z.
  • Data-use warning: Inclusive long-context boundary follows the explicit pricing table and fresh pricing.md, despite generic model/cache prose saying 'exceed'.
  • Long-context rate at or above 200,000 input tokens: Input $4.00/M; cached input $1.00/M; output $12.00/M.
  • Documented batch rate: Input $1.60/M; cached input $0.40/M; output $4.80/M.
  • Documented long-context batch rate: Input $3.20/M; cached input $0.80/M; output $9.60/M.
  • Documented US inference multiplier: 1.1×.
xAIGrok 4.5FrontierVerify provider price ↗$2.00$0.30$6.00500,000 tokens
  • Exact public API model: grok-4.5. Coding/engineering and reasoning model with a genuinely distinct cached-input rate from Grok 4.6/4.7. Aliases grok-4.5-latest and grok-build-latest are not separate priced rows. Batch supported with official 20% discount. Not currently served by the US endpoint. Global/default standard text scope; shared cache conditions apply.
  • Supplier API identity: grok-4.5.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:34Z.
  • Data-use warning: Inclusive long-context boundary follows the explicit pricing table and fresh pricing.md, despite generic model/cache prose saying 'exceed'.
  • Long-context rate at or above 200,000 input tokens: Input $4.00/M; cached input $0.60/M; output $12.00/M.
  • Documented batch rate: Input $1.60/M; cached input $0.24/M; output $4.80/M.
  • Documented long-context batch rate: Input $3.20/M; cached input $0.48/M; output $9.60/M.
xAIGrok 4.3Balanced · 1M contextVerify provider price ↗$1.25$0.20$2.501,000,000 tokens
  • Exact public API model: grok-4.3; grok-4.3-latest is an alias, not another row. Global/default standard text prices and existing batch values verified. Supports configurable reasoning including none. The official batch discount is 20%, not 50%; no expiry is stated. Removed baseline usResidencyMultiplier because the current US endpoint serves only grok-4.7 and grok-4.6. Shared cache conditions apply.
  • Supplier API identity: grok-4.3.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:34Z.
  • Data-use warning: Grok 4.3 is not currently available on the US endpoint; do not apply its old 1.1 residency multiplier. Inclusive long-context boundary follows explicit pricing table and fresh pricing.md despite generic model/cache prose saying 'exceed'.
  • Long-context rate at or above 200,000 input tokens: Input $2.50/M; cached input $0.40/M; output $5.00/M.
  • Documented batch rate: Input $1.00/M; cached input $0.16/M; output $2.00/M.
  • Documented long-context batch rate: Input $2.00/M; cached input $0.32/M; output $4.00/M.
xAIGrok Build 0.1BudgetVerify provider price ↗$1.00$0.20$2.00256,000 tokens
  • Exact public API model: grok-build-0.1, an agentic coding/engineering text model, not the Grok Build subscription product. Model page documents aliases grok-code-fast-1, grok-code-fast and grok-code-fast-1-0825; they are not separately priced original models. Batch API is explicitly not supported, so no batch price is included. Not currently served by the US endpoint. Shared cache conditions apply.
  • Supplier API identity: grok-build-0.1.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:34Z.
  • Data-use warning: Inclusive long-context boundary follows explicit pricing table and fresh pricing.md despite generic model/cache prose saying 'exceed'. Public HTML contains an internal batchDiscountPercent value for this model, but visible pricing excludes it and model Details says Batch API Not supported; no batch discount is applied.
  • Long-context rate at or above 200,000 input tokens: Input $2.00/M; cached input $0.40/M; output $4.00/M.
xAIGrok 4.20 (Reasoning)BalancedVerify provider price ↗$1.25$0.20$2.501,000,000 tokens
  • Current stable public API alias: grok-4.20-reasoning, explicitly documented as an alias of priced model grok-4.20-0309-reasoning. One row represents this reasoning variant; dated snapshots and latest aliases are not additional rows. Global/default standard text prices; reasoning tokens billed at output rate. Batch supported with official 20% discount. Not currently served by the US endpoint. Shared cache conditions apply.
  • Supplier API identity: grok-4.20-reasoning.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:34Z.
  • Data-use warning: Inclusive long-context boundary follows explicit pricing table and fresh pricing.md despite generic model/cache prose saying 'exceed'.
  • Long-context rate at or above 200,000 input tokens: Input $2.50/M; cached input $0.40/M; output $5.00/M.
  • Documented batch rate: Input $1.00/M; cached input $0.16/M; output $2.00/M.
  • Documented long-context batch rate: Input $2.00/M; cached input $0.32/M; output $4.00/M.
xAIGrok 4.20 (Non-Reasoning)BalancedVerify provider price ↗$1.25$0.20$2.501,000,000 tokens
  • Current stable public API alias: grok-4.20-non-reasoning, explicitly documented as an alias of priced model grok-4.20-0309-non-reasoning. Distinct non-reasoning model variant; no duplicate dated/latest aliases included. Global/default standard text prices; batch supported with official 20% discount. Not currently served by the US endpoint. Shared cache conditions apply.
  • Supplier API identity: grok-4.20-non-reasoning.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:34Z.
  • Data-use warning: Inclusive long-context boundary follows explicit pricing table and fresh pricing.md despite generic model/cache prose saying 'exceed'. The model page also displays a generic reasoning capability label; the exact named non-reasoning identity is retained without relabeling it as the reasoning model.
  • Long-context rate at or above 200,000 input tokens: Input $2.50/M; cached input $0.40/M; output $5.00/M.
  • Documented batch rate: Input $1.00/M; cached input $0.16/M; output $2.00/M.
  • Documented long-context batch rate: Input $2.00/M; cached input $0.32/M; output $4.00/M.
xAIGrok 4.20 Multi-Agent BetaBalanced · BetaVerify provider price ↗$1.25$0.20$2.501,000,000 tokens
  • Current stable public API alias: grok-4.20-multi-agent, documented as alias of priced model grok-4.20-multi-agent-0309 and used by the official guide. This is a distinct publicly priced beta multi-agent text/reasoning tier, not multiple snapshot rows. Works without built-in tools through Responses API or xAI SDK; Chat Completions is not supported. All leader and sub-agent input, output and reasoning tokens are billed, not only visible final answer tokens. Agent counts change usage, not the published per-token rate. Batch supported per model Details with 20% official discount; unavailable under ZDR. Not currently served by the US endpoint. Shared cache conditions apply.
  • Supplier API identity: grok-4.20-multi-agent.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:34Z.
  • Data-use warning: Beta interface may change; hidden sub-agent token use can materially increase total cost. Tool invocations, if enabled, cost extra and are excluded from these text rates. Inclusive long-context boundary follows explicit pricing table and fresh pricing.md despite generic model/cache prose saying 'exceed'.
  • Long-context rate at or above 200,000 input tokens: Input $2.50/M; cached input $0.40/M; output $5.00/M.
  • Documented batch rate: Input $1.00/M; cached input $0.16/M; output $2.00/M.
  • Documented long-context batch rate: Input $2.00/M; cached input $0.32/M; output $4.00/M.
KimiKimi K3.1 ProFrontier · 1M contextVerify provider price ↗$2.00$0.20$10.001,048,576 tokens
  • Stable API identity: kimi-k3.1-pro. Direct global/default text API, pay-as-you-go, before taxes. Available after a successful minimum USD 1 top-up; not a membership or Kimi Code plan. Flat prices throughout the context window, with no long-context surcharge. Thinking is always enabled; output usage includes reasoning. Implicit prefix caching defaults to 5m TTL: cache writes cost USD 2.00 per 1M tokens for 5m or USD 4.00 for 1h. Batch writes cost USD 1.20 for 5m or USD 2.40 for 1h per 1M tokens. Cache reads, writes and uncached remainder are mutually exclusive input categories, not additive charges on the same tokens; hits renew the original TTL without another write fee.
  • Supplier API identity: kimi-k3.1-pro.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:24Z.
  • Data-use warning: The supplied USD 1.40 input / 7.00 output / 0.14 cache-hit offer is unverified and is not used. Cache-hit pricing alone does not price cache writes; 1h writes need separate calculation. Batch caching is best-effort, not guaranteed.
  • Documented batch rate: Input $1.20/M; cached input $0.12/M; output $6.00/M.
KimiKimi K3.1 Pro HighspeedFrontier · Fast / 1M contextVerify provider price ↗$4.00$0.20$20.001,048,576 tokens
  • Stable API identity: kimi-k3.1-pro-highspeed. Separately priced speed tier, same intelligence as K3.1 Pro; official model guide reports about 2–3 times the speed. USD global/default direct text API rates before taxes; minimum USD 1 top-up required. Flat context pricing. Implicit cache-write rates: USD 4.00 per 1M tokens for default 5m TTL and USD 8.00 for 1h. Cached input is USD 0.20, not twice the standard variant's cache-hit rate. Cache reads, writes and uncached tokens are mutually exclusive categories. No batch rate is listed for this identity.
  • Supplier API identity: kimi-k3.1-pro-highspeed.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:24Z.
  • Data-use warning: Do not double an entire standard-model bill to estimate Highspeed: the cache-hit rate remains unchanged, and cache writes require TTL-specific accounting. No unsupported batch rate or residency multiplier has been inferred.
KimiKimi K3Frontier · 1M contextVerify provider price ↗$3.00$0.30$15.001,048,576 tokens
  • Baseline retained: all three published token rates and context remain documented. Stable API identity: kimi-k3, previous-generation flagship still available through direct global/default text API after minimum USD 1 top-up. No context-length pricing tiers. Thinking is always on and contributes to output usage. Cache writes are USD 3.00 per 1M tokens for default 5m TTL or USD 6.00 for 1h; batch writes USD 1.80/3.60 respectively. Cache-hit tokens do not also incur a write charge. Cache reads, writes and uncached remainder are mutually exclusive input categories.
  • Supplier API identity: kimi-k3.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:24Z.
  • Data-use warning: Cache-hit rates alone do not price TTL writes; do not also charge regular input on the same cache-write tokens. Batch cache hits are best-effort.
  • Documented batch rate: Input $1.80/M; cached input $0.18/M; output $9.00/M.
KimiKimi K2.6Legacy · Existing users onlyVerify provider price ↗$0.95$0.16$4.00262,144 tokens
  • Baseline retained as documented legacy, not available to newly registered users from October 31, 2026; existing users remain eligible. Stable API identity is kimi-k2.6; baseline calculator id is preserved. General-purpose text/coding/reasoning model; thinking defaults on but may be disabled, with no separately priced mode row. USD global/default non-batch rates before taxes. K2.6 does not support separately billed cache writes. Official batch table explicitly lists cached input USD 0.10 per 1M tokens; no multiplication of a percentage was used.
  • Supplier API identity: kimi-k2.6.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:24Z.
  • Data-use warning: Supplier plans to retire all K2 series models by the end of December 2026. Preserve only for eligible existing-user estimates; not a new-user model recommendation. Batch cache hits are best-effort. Do not substitute the mathematically derived 0.096 for the published batch cache rate 0.10.
  • Documented batch rate: Input $0.57/M; cached input $0.10/M; output $2.40/M.
KimiKimi K2.7 CodeLegacy · Code / existing users onlyVerify provider price ↗$0.95$0.19$4.00262,144 tokens
  • Baseline retained as documented existing-user legacy. Stable API identity: kimi-k2.7-code; baseline calculator id preserved. Direct global/default text/coding API rates, USD per 1M tokens before taxes. Thinking and Preserved Thinking are always on; reasoning contributes to output and historical reasoning sent again contributes to input. No separately billed cache writes. Batch rates are explicit published table entries.
  • Supplier API identity: kimi-k2.7-code.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:24Z.
  • Data-use warning: Unavailable to newly registered users from October 31, 2026. All K2 series planned for retirement by end December 2026. Batch cache hits are best-effort. Caching guide FAQ abbreviates K2.7 identities as kimi-k2.7/kimi-k2.7-highspeed; model list and pricing table use the full -code identities, which are the identities validated here.
  • Documented batch rate: Input $0.57/M; cached input $0.114/M; output $2.40/M.
KimiKimi K2.7 Code HighSpeedLegacy · Fast code / existing users onlyVerify provider price ↗$1.90$0.38$8.00262,144 tokens
  • Baseline retained as documented existing-user legacy and genuinely distinct priced speed tier. Stable API identity: kimi-k2.7-code-highspeed; baseline calculator id preserved. USD global/default direct text/coding API rates before taxes. Same model intelligence and thinking behavior as K2.7 Code; supplier reports approximately 180 output tokens/s, up to 260 in short contexts. No separately billed cache writes. No batch rate listed for this identity.
  • Supplier API identity: kimi-k2.7-code-highspeed.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:24Z.
  • Data-use warning: Unavailable to newly registered users from October 31, 2026; K2 series planned for retirement by end December 2026. Do not infer batch pricing from the standard K2.7 Code tier.
Meta MuseMuse Wonder 1.0 StandardFrontier · Preview · StandardVerify provider price ↗$4.00$0.20$20.001,048,576 tokens
  • Exact API identity: muse-wonder-1.0. Public Meta Model API pay-as-you-go preview, Standard mode only. Direct default non-batch text rates in USD per 1M tokens. Standard content is not used to train Meta models; this is not zero data retention. No long-context premium. Cache-hit rate applies only to reported cached input tokens; automatic cache misses/first prefixes consume uncached input. Separate cache-write or retention/storage rates are not published in the reviewed pricing/cache docs; do not treat the cache-hit rate as a cache-write price. Default in-memory retention has no fixed documented TTL; optional Responses 24h retention is an evictable hint. Model docs say unavailable in South Korea. Not Muse Wonder Full.
  • Supplier API identity: muse-wonder-1.0.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:00Z.
Meta MuseMuse Spark 1.3 StandardBalanced · StandardVerify provider price ↗$1.25$0.15$4.251,048,576 tokens
  • Exact API identity: muse-spark-1.3; retained calculator ID is not the API ID. Latest Spark version, recommended for new work. Standard-tier prompts and completions are not used to train Meta models; Standard is not ZDR. Direct default non-batch text USD rates per 1M tokens. No long-context premium. Automatic prefix cache-hit pricing only; uncached tokens use input pricing. Separate cache-write/storage prices are not published in reviewed docs. Default in-memory cache has no fixed published TTL; optional Responses 24h retention is a hint. Supports max reasoning unlike its Contributor variant.
  • Supplier API identity: muse-spark-1.3.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:00Z.
Meta MuseMuse Spark 1.2 StandardLegacy · StandardVerify provider price ↗$1.25$0.15$4.251,048,576 tokens
  • Exact API identity: muse-spark-1.2; retained calculator ID is not the API ID. Previous Spark version, still explicitly documented and publicly priced on Meta Model API; Legacy denotes previous generation, not retired. Standard-tier prompts and completions are not used to train Meta models; not ZDR. Direct default non-batch text USD per 1M tokens; no long-context premium. Cache-hit rate only; cache misses use uncached input. Cache-write/storage price not separately published. Default in-memory eviction has no fixed documented TTL; optional Responses 24h retention is a hint.
  • Supplier API identity: muse-spark-1.2.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:00Z.
Meta MuseMuse Spark 1.1 StandardLegacy · StandardVerify provider price ↗$1.25$0.15$4.251,048,576 tokens
  • Exact API identity: muse-spark-1.1; retained calculator ID is not the API ID. Original Spark version remains publicly documented and priced on Meta Model API, Standard only; Legacy does not mean retired. Not available in Muse Code. Standard-tier prompts and completions are not used to train Meta models; not ZDR. Direct default non-batch text USD per 1M tokens, no long-context premium. Automatic cache-hit rate only; uncached input is priced normally. No separately published cache-write/storage rate or fixed default TTL in reviewed docs; optional Responses 24h retention is only a hint.
  • Supplier API identity: muse-spark-1.1.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:00Z.
Meta MuseMuse Spark 1.3 ContributorBudget · ContributorVerify provider price ↗$0.10$0.002$0.201,048,576 tokens
  • Exact API identity: muse-spark-1.3-contributor; retained calculator ID is not the API ID. Public direct pay-as-you-go Contributor mode; official supplier conditional rates, not an assumed reseller discount. No percentage discount or expiry supplied in baseline; current docs state absolute rates without a dated promotion expiry. Direct default non-batch text USD per 1M tokens, no long-context premium. Same Spark 1.3 model but max reasoning is unavailable in Contributor mode. Automatic cache-hit pricing only; uncached tokens billed at input rate. Separate cache-write/storage rates not published; default in-memory TTL not fixed, optional Responses 24h is a hint.
  • Supplier API identity: muse-spark-1.3-contributor.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:00Z.
  • Data-use warning: Contributor pricing permits Meta to use prompts and completions to train future Meta models. This also covers submitted code, files and context. Do not submit sensitive, confidential or personal information, including confidential software code. Previously Standard content reused in a Contributor request becomes Contributor content. Contributor variants are unavailable to ZDR teams and offer no per-traffic training opt-out.
Meta MuseMuse Spark 1.2 ContributorLegacy · ContributorVerify provider price ↗$0.10$0.002$0.201,048,576 tokens
  • Exact API identity: muse-spark-1.2-contributor; retained calculator ID is not the API ID. Previous Spark version remains documented and publicly priced; Legacy denotes previous generation, not retired. Official supplier Contributor conditional rates, not an assumed reseller discount. No percentage offer or expiry supplied; current absolute rates have no documented promotion expiry. Direct default non-batch text USD per 1M tokens; no long-context premium. Cache-hit pricing only, misses billed at input price. Separate cache-write/storage prices not published; default in-memory cache has no fixed TTL, optional Responses 24h is a hint.
  • Supplier API identity: muse-spark-1.2-contributor.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:00Z.
  • Data-use warning: Contributor pricing permits Meta to use prompts and completions to train future Meta models. This also covers submitted code, files and context. Do not submit sensitive, confidential or personal information, including confidential software code. Previously Standard content reused in a Contributor request becomes Contributor content. Contributor variants are unavailable to ZDR teams and offer no per-traffic training opt-out.
DeepSeekDeepSeek V4.1 FlashBudget · Peak/off-peak pricingVerify provider price ↗$0.30$0.006$1.20Not listed
  • Current stable public API identity is deepseek-flash, served by DeepSeek-V4.1-Flash-1017. Direct standard non-batch text rates in USD per 1M tokens apply to both thinking and non-thinking modes; thinking is enabled by default. rate contains the published PEAK schedule as a conservative untimed calculator baseline, not an all-hours flat price. Official OFF-PEAK rates are input/cache-miss $0.15, output $0.60, and cached-input/cache-hit $0.003 per 1M tokens. Peak hours are 01:00-04:00 and 06:00-10:00 UTC Monday-Friday, excluding Chinese public holidays; every other hour, all weekends and Chinese public holidays are off-peak. These schedules apply to API users generally and are not a reseller promotion. The documentation states 1M total context and a maximum output of 384K (393216 tokens); contextTokens is omitted because the checked public pages do not give an exact integer for the 1M context limit. Automatic disk-prefix caching is enabled for all users. Cache-hit pricing applies only to fully matching persisted prefix units; cache misses, including initial uncached requests, use the input rate. The current cache guide describes persistence at user-input/model-output boundaries, detected shared prefixes and fixed intervals, best-effort hits, construction taking seconds, and unused-cache eviction usually within hours to days rather than a guaranteed TTL. The historical official cache announcement says there is no additional cache-service fee and storage is free; the current pricing page publishes no separate cache-write or storage charge. Its historical token prices and 64-token storage rule are not used as current pricing or a current minimum. No documented long-context token-price uplift, batch schedule, fast surcharge, or US-residency multiplier was found in the current pricing page; those fields are omitted. This row covers text only, not image input, search/fetch tools, or storage products.
  • Supplier API identity: deepseek-flash.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:33Z.
  • Data-use warning: Do not display the peak rate as an all-hours flat price or the off-peak rate as an unconditional list price. Preserve both time schedules and Chinese-holiday eligibility. No supplier confirmation was found for the supplied additional discounted rates. Cache-hit rate alone must not be interpreted as a premium cache-write price or guaranteed cache availability.
  • Conservative default: peak. Peak: Monday-Friday, 01:00-04:00 and 06:00-10:00 UTC, excluding Chinese public holidays. Off-peak: All other hours, all weekends, and Chinese public holidays. Optional off-peak rate: Input $0.15/M; cached input $0.003/M; output $0.60/M. Check the supplier's time and holiday rules; the calculator does not infer the current billing window.
DeepSeekDeepSeek V4 ProBalanced · Peak/off-peak pricingVerify provider price ↗$1.32$0.044$3.96Not listed
  • Current stable public API identity deepseek-v4-pro is served by DeepSeek-V4-Pro-0813. Direct standard non-batch text rates in USD per 1M apply to both thinking and non-thinking modes. rate contains the published PEAK schedule as a conservative untimed calculator baseline, not an all-hours flat price. Official OFF-PEAK rates are input/cache-miss $0.66, output $1.98, and cached-input/cache-hit $0.022 per 1M tokens. Peak hours are 01:00-04:00 and 06:00-10:00 UTC Monday-Friday, excluding Chinese public holidays; every other hour, weekends and Chinese public holidays are off-peak. The pricing page states 1M context and 384K maximum output; the chat reference gives maximum output as 393216 and defines context as input plus output. contextTokens is omitted because the checked pages do not specify an exact integer for 1M context. Caching is automatic, disk-based, best-effort and dependent on full persisted-prefix-unit matches; initial/missed input uses the cache-miss rate. Construction takes seconds and idle caches generally expire after hours to days, with no guaranteed TTL. Historical first-party cache documentation says no extra cache-service fees and free storage; the current pricing page lists no separate cache-write/storage premium. No documented long-context token-price uplift, batch schedule, fast surcharge or US-residency multiplier is published on the current price page; those fields are omitted. Pro is currently documented as an active service, not kept merely as a stale legacy row.
  • Supplier API identity: deepseek-v4-pro.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:33Z.
  • Data-use warning: The September 10 release article still contains a planned September 14 Pro reroute, but the official changelog explicitly reverses that plan and says V4 Pro service and billing continue after September 14. Current pricing, first-call and chat API reference pages independently continue to list deepseek-v4-pro. Prefer those current documents. Do not mistake the time-based schedules for long-context or fast tiers.
  • Conservative default: peak. Peak: Monday-Friday, 01:00-04:00 and 06:00-10:00 UTC, excluding Chinese public holidays. Off-peak: All other hours, all weekends, and Chinese public holidays. Optional off-peak rate: Input $0.66/M; cached input $0.022/M; output $1.98/M. Check the supplier's time and holiday rules; the calculator does not infer the current billing window.
Z.AIGLM-5.3Frontier · Latest flagshipVerify provider price ↗$1.40$0.26$4.401,000,000 tokens
  • Standard direct API, non-batch text scope. Text-only input and text output; reasoning is always enabled, with low/high/max effort. Context is published as 1M, a marketed capacity rather than a separately documented exact integer boundary. Implicit cache: new content is charged at standard input rates; the listed cache rate applies to cache-hit tokens. Cached Input Storage is 'Limited-time Free'; no expiry date, numerical TTL, guaranteed cache lifetime, minimum cache length, or separate cache-write surcharge is published in the checked pricing/cache docs. This does not establish permanently free cache writes/storage. No automatic long-context surcharge, batch discount, or residency multiplier is documented.
  • Supplier API identity: glm-5.3.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:21:25Z.
Z.AIGLM-5.3-FlashBudget · Requested model verifiedVerify provider price ↗$0.15$0.03$0.501,000,000 tokens
  • Exact requested model, not GLM-5.3 or GLM-5.3-FlashX. A native multimodal model with an explicitly supported plain-text API mode and text output; rates here cover text only, not image/video/file processing. Docs publish 1M context and mandatory reasoning. Standard prices are retained; supplied 0.075/0.25/0.015 discounted rates are not verified as a currently eligible direct API offer. Implicit cache new content uses standard input billing; cached-input price is for hits, not a full cache-write/storage price. Storage is currently described as 'Limited-time Free', without an end date or numerical TTL. No documented default long-context price trigger.
  • Supplier API identity: glm-5.3-flash.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:21:25Z.
  • Data-use warning: Do not apply an assumed 50% discount. Supplier blog's '$0.045 per task (discounted)' benchmark is not a per-token rate or current promotion contract. Coding Plan off-peak 50% points usage is not standard API token pricing.
Z.AIGLM-5.3-FlashXBalanced · High-speed tierVerify provider price ↗$0.37$0.075$1.251,000,000 tokens
  • Distinct named public API tier, not a guessed speed multiplier on Flash. Supplier says now live with inference speed of 200 tokens/s; this is a performance claim, not a guarantee. Plain-text requests and text output are supported; only text rates are captured. Published context 1M. New cache content uses normal input billing; cache-hit rate does not price all writes/storage. Storage Limited-time Free with no documented expiry or numerical TTL. No default long-context surcharge documented. Represented as its own API model, not duplicated as fastRate on Flash.
  • Supplier API identity: glm-5.3-flashx.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:21:25Z.
Z.AIGLM-5.3-FlashX-LegalBalancedVerify provider price ↗$0.37$0.075$1.251,000,000 tokens
  • Supplier-managed, separately named and priced reasoning/text API model optimized for legal research, contract review and drafting; not a customer fine-tuning service or enterprise quote. Included as a distinct active GLM-5.3 family identity. Optional external legal MCP integrations are not included in these token rates. Plain-text scope only. Published context 1M. Cache new content uses standard input billing; hits use listed cached-input rate; storage is Limited-time Free with unspecified expiry and numerical TTL.
  • Supplier API identity: glm-5.3-flashx-legal.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:21:25Z.
Z.AIGLM-5.2LegacyVerify provider price ↗$1.40$0.26$4.401,000,000 tokens
  • Still explicitly documented and in the current direct API enum, not retired. Legacy denotes a predecessor to GLM-5.3, not unsupported status. Stable model code retained, not a dated snapshot alias. Text-only; published context 1M. Implicit-cache new content uses standard input rate; cache hits use listed rate. Storage Limited-time Free, no published numerical TTL/expiry. No documented long-context price step.
  • Supplier API identity: glm-5.2.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:21:25Z.
Z.AIGLM-5.1LegacyVerify provider price ↗$1.40$0.26$4.40200,000 tokens
  • Still explicitly documented and in the current direct API enum, not retired. Legacy is a relative-generation label. Stable API identity, not a snapshot alias. Text-only; docs advertise 200K context, represented here as 200000 marketed tokens, not a claim of an exact 202752/204800 boundary. Cache new content is standard input; hits use listed rate. Storage Limited-time Free, unspecified expiry and numerical TTL.
  • Supplier API identity: glm-5.1.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:21:25Z.
Z.AIGLM-5LegacyVerify provider price ↗$1.00$0.20$3.20200,000 tokens
  • Still explicitly documented and in the current direct API enum, not retired. Stable GLM-5 identity; not substituted for GLM-5-Turbo. Text-only; context published as 200K marketed tokens. Cache new content uses standard input and cache hits use listed rate; storage Limited-time Free, no published numerical TTL or expiry.
  • Supplier API identity: glm-5.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:21:25Z.
Z.AIGLM-4.7LegacyVerify provider price ↗$0.60$0.11$2.20200,000 tokens
  • Still explicitly documented and in the current direct API enum, not retired. Exact GLM-4.7 identity, not its Flash/FlashX variants. Text-only; docs publish 200K context. Cache new content standard input; hits listed cached-input rate. Storage Limited-time Free, numerical TTL and expiry unspecified.
  • Supplier API identity: glm-4.7.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:21:25Z.
Z.AIGLM-4.6LegacyVerify provider price ↗$0.60$0.11$2.20200,000 tokens
  • Still explicitly documented and in the current direct API enum, not retired. Text-only; published 200K context. Cache new content standard input; cache hits listed cached-input rate. Storage Limited-time Free, unspecified numerical TTL and expiry. Legacy does not mean unavailable.
  • Supplier API identity: glm-4.6.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:21:25Z.
TencentHunyuan Hy4 PreviewFrontier · PreviewVerify provider price ↗$0.834$0.042$2.501Not listed
  • Exact public API parameter is hy4-preview. Tencent identifies it as an early Hunyuan Hy4 iteration, not Hy3, an older Hunyuan Turbo model, or a GA Hy4 release. Scope is Tencent-operated TokenHub pay-as-you-go, standard online non-batch text requests using the Singapore endpoint with Global resource scheduling. Documentation renders context as 1M, maximum input as 960k and maximum output as 64k; numeric contextTokens is omitted because these abbreviated limits are not an exact integer specification. Supports Chat Completions, Responses and Anthropic Messages after account/model activation. No input-length pricing condition or peak/off-peak condition is listed for this model.
  • Supplier API identity: hy4-preview.
  • Deployment scope: Singapore / Global.
  • Source checked: 2026-11-02T22:18:32Z.
  • Data-use warning: Preview is not GA. Rates are USD per million billed tokens, not the mainland release's CNY prices. Cached-input rate covers actual cache hits only. The pricing table shows '-' for both 5-minute and 1-hour cache-write prices; this is not an explicit zero-price statement. HY-specific cache-write, storage and exact TTL charges/limits are not published in the reviewed documentation, so no free writes/storage, fixed TTL, guaranteed hit or minimum cacheable-prefix length is assumed. Account activation is required; the billing guide states USD 10 is frozen per model when enabling postpaid or claiming trial quota. Trial/subscription benefits are not applied to these rates.
TencentHunyuan Hy3BalancedVerify provider price ↗$0.132$0.033$0.528Not listed
  • Current official-version API identity is hy3, not hy3-preview. Tencent describes general-purpose, coding, long-text, reasoning and agent capabilities. Scope is Tencent-operated TokenHub pay-as-you-go standard online non-batch text requests at the Singapore/Global endpoint. Model documentation renders context as 256k, maximum input as 192k and maximum output as 128k; exact integer contextTokens is omitted rather than inferred from abbreviations. Supports Chat Completions, Responses and Anthropic Messages after activation. No length-tier or peak/off-peak condition is listed; no batch/fast rate is asserted.
  • Supplier API identity: hy3.
  • Deployment scope: Singapore / Global.
  • Source checked: 2026-11-02T22:18:32Z.
  • Data-use warning: Cache price is actual cached-input consumption only. Both 5-minute and 1-hour cache-write cells are '-' rather than numeric zero. Reviewed HY docs do not publish an exact cache TTL, storage price, minimum eligible prompt length or separately quantified write charge; do not assume free cache writes/storage or guaranteed hits. Account/model activation and the billing guide's USD 10 frozen amount apply; free-trial and token-plan credits are not standard token-rate discounts.
MistralMistral Large 4Frontier · Public PreviewVerify provider price ↗$0.70$0.07$2.10Not listed
  • Direct Mistral global/default Standard Tier text-token prices, USD per 1 million tokens. Card displays context as 1M; exact integer limit not independently documented in the material inspected, so contextTokens is omitted. One documented API identity is retained; latest/snapshot aliases are not separate rows. Cache-hit price is not a cache-write or storage price; see supplier sourceNotes.
  • Supplier API identity: mistral-large-4.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:14Z.
  • Data-use warning: Public Preview permits silent updates. The marketing pricing FAQ's unversioned Mistral Large example is not the Large 4 price. No automatic long-context premium is documented in the inspected rate table.
MistralMistral Small 4Balanced · GAVerify provider price ↗$0.15$0.015$0.60Not listed
  • General text, reasoning and coding in a single model. Direct Mistral global/default Standard Tier USD per 1 million text tokens. Model card displays 256k context; exact integer not established, so contextTokens is omitted. Cache eligibility and undisclosed write/TTL/storage details are in sourceNotes.
  • Supplier API identity: mistral-small-2603.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:14Z.
MistralMinistral 3 14BBudget · GAVerify provider price ↗$0.20$0.02$0.20Not listed
  • Distinct 14B tier of Ministral 3. Direct Mistral global/default Standard Tier text rates in USD per 1 million tokens; not self-hosting cost. Card displays 256k context; exact integer not established. See sourceNotes for cache conditions.
  • Supplier API identity: ministral-14b-2512.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:14Z.
MistralMinistral 3 8BBudget · GAVerify provider price ↗$0.15$0.015$0.15Not listed
  • Distinct 8B tier of Ministral 3. Direct Mistral global/default Standard Tier text rates in USD per 1 million tokens. Card displays 256k context; exact integer not established. This is not retired Ministral 8B ministral-8b-2410. See sourceNotes for cache conditions.
  • Supplier API identity: ministral-8b-2512.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:14Z.
MistralMinistral 3 3BBudget · GAVerify provider price ↗$0.10$0.01$0.10Not listed
  • Distinct 3B tier of Ministral 3. Direct Mistral global/default Standard Tier text rates in USD per 1 million tokens. Card displays 256k context; exact integer not established. This is not retired Ministral 3B ministral-3b-2410. See sourceNotes for cache conditions.
  • Supplier API identity: ministral-3b-2512.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:20Z.
MistralCodestral 25.08Balanced · GAVerify provider price ↗$0.30$0.03$0.90Not listed
  • Current Codestral code-generation/completion API, not Codestral Embed or retired Codestral 25.01. Direct Mistral global/default Standard Tier USD per 1 million text tokens. Card displays 128k context; exact integer not established. See sourceNotes for shared-prefix/FIM cache conditions.
  • Supplier API identity: codestral-2508.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:20Z.
MistralZ.ai GLM 5.3 (Mistral-hosted)Frontier · GA · Third-partyVerify provider price ↗$1.40$0.14$4.40Not listed
  • Z.ai model hosted by Mistral without modifications; provider identifies the direct API supplier, not model authorship. USD per 1 million text tokens on Mistral's global/default Standard Tier, not Z.ai's own API pricing. Card displays 1M context and 128k max output; exact integer limits not established. See sourceNotes for cache conditions.
  • Supplier API identity: zai-glm-5-3.
  • Deployment scope: Supplier default.
  • Model author: Z.AI. API supplier: Mistral.
  • Source checked: 2026-11-02T22:19:16Z.
MistralZ.ai GLM 5.3 Flash (Mistral-hosted)Budget · GA · Third-partyVerify provider price ↗$0.15$0.03$0.50Not listed
  • Distinct Flash tier, Z.ai model hosted without Mistral modifications. Mistral direct global/default Standard Tier text rates in USD per 1 million tokens, not Z.ai API rates. Card displays 1M context and 128k max output; exact integer limits not established.
  • Supplier API identity: zai-glm-5-3-flash.
  • Deployment scope: Supplier default.
  • Model author: Z.AI. API supplier: Mistral.
  • Source checked: 2026-11-02T22:19:17Z.
  • Data-use warning: Model-specific published cached input is $0.03/M, not the blanket 10%-of-input statement in Mistral's caching guide. Use the explicit current model card/table; no inferred discount. Cache-write, storage and TTL charges are not separately specified in inspected documentation.
MistralDeepSeek V4.1 Flash 1017 (Mistral-hosted)Balanced · Public Preview · Third-partyVerify provider price ↗$0.30$0.007$1.20Not listed
  • DeepSeek model hosted by Mistral without modifications. Exact card/API identity contains v4-1 although documentation URL contains v4; not relabeled as another DeepSeek model. One currently offered identity is retained, not a snapshot catalog. Direct Mistral global/default Standard Tier text-token USD rates per 1 million tokens. Card displays 1M context and 128k max output; exact integer limits not established.
  • Supplier API identity: deepseek-v4-1-flash-1017.
  • Deployment scope: Supplier default.
  • Model author: DeepSeek. API supplier: Mistral.
  • Source checked: 2026-11-02T22:19:20Z.
  • Data-use warning: Public Preview permits silent updates. Explicit cached input $0.007/M differs from the generic caching guide's 10%-of-input statement; exact current model card/table wins. No assumed cache-write/storage rate or TTL is added.
AlibabaQwen3.8 Max (International)Frontier · Thinking + codingVerify provider price ↗$2.00$0.25$6.001,000,000 tokens
  • API model=qwen3.8-max; Singapore endpoint, International inference scope excluding Chinese mainland. Standard text-only rates; same output price for thinking and non-thinking, with chain-of-thought tokens billed as output. No prompt-price tiers. cachedInputPerM is implicit cache; explicit cache read is USD 0.17/M and creation USD 2.5/M. Context is 1,000,000; maximum input 991,808 non-thinking / 983,616 thinking. Text-only batch is not supported in this scope.
  • Supplier API identity: qwen3.8-max.
  • Deployment scope: International.
  • Source checked: 2026-11-02.
AlibabaQwen3.8 Max (Global)Frontier · Global scopeVerify provider price ↗$1.65$0.206$4.9511,000,000 tokens
  • Calculator id is scope-qualified; API model remains qwen3.8-max, not qwen3.8-max-global. Global scope at Frankfurt, Virginia, Tokyo or Hong Kong; inference may execute in Chinese mainland or overseas. Same standard thinking/non-thinking text output price. No prompt-price tiers. cachedInputPerM is implicit cache; explicit read USD 0.137/M, creation USD 2.063/M. Maximum input 991,808 non-thinking / 983,616 thinking. Global batch unsupported. US-only residency is a different scope priced at USD 2 input / 6 output / 0.25 implicit cache per M; no single exact residency multiplier is assumed.
  • Supplier API identity: qwen3.8-max.
  • Deployment scope: Global.
  • Source checked: 2026-11-02.
AlibabaQwen3.7 Plus (International)Balanced · Thinking + codingVerify provider price ↗$0.40$0.08$1.601,000,000 tokens
  • API model=qwen3.7-plus, currently equivalent to qwen3.7-plus-2026-05-26; Singapore / International. Listed standard prices exclude limited-time 20% off labels. Tier selected by TOTAL INPUT tokens per request and applies to all input and output tokens. Same output price in thinking and non-thinking; reasoning tokens billed as output. Implicit cache is in rate; explicit cache creation/read are USD 0.5/0.04 per M at input≤256,000 and 1.5/0.12 per M above 256,000. Maximum input 991,808 / thinking 983,616. Batch File API confirms text support and 50% of realtime price, but batch context is capped at 256K; optional batch rates omitted to keep standard/non-batch scope separate.
  • Supplier API identity: qwen3.7-plus.
  • Deployment scope: International.
  • Source checked: 2026-11-02.
  • Data-use warning: Official pricing table confirms a limited-time 20% off label, but no expiry or complete eligibility is established from public documents. No promotional discount applied.
  • Long-context rate above 256,000 input tokens: Input $1.20/M; cached input $0.24/M; output $4.80/M.
AlibabaQwen3.7 Plus (Global)Balanced · Global scopeVerify provider price ↗$0.276$0.056$1.1011,000,000 tokens
  • API model=qwen3.7-plus (calculator id is scope-qualified); Global at Frankfurt/Virginia/Tokyo/Hong Kong, including mainland inference resources. Total-input tier applies to all tokens. Both thinking/non-thinking output modes use the listed output rate. Use published rounded cache prices rather than multiplying the general cache ratio. Explicit creation/read USD 0.344/0.028 per M at input≤256K and 1.032/0.083 per M above 256K. Maximum input 991,808 / thinking 983,616. Global batch unsupported. US-only scope separately lists standard USD 0.4/1.6/0.08 per M at input≤256K and 1.2/4.8/0.24 above; not represented by an assumed multiplier.
  • Supplier API identity: qwen3.7-plus.
  • Deployment scope: Global.
  • Source checked: 2026-11-02.
  • Data-use warning: Official Global list-price table labels limited-time night 60% off / daytime 20% off, with night 22:00–08:00 UTC+8 based on billing time. Expiry and full eligibility are not established; no discounts applied.
  • Long-context rate above 256,000 input tokens: Input $0.826/M; cached input $0.166/M; output $3.301/M.
AlibabaQwen3.8 Flash (International)Budget · Thinking + codingVerify provider price ↗$0.15$0.016$0.471,000,000 tokens
  • API model=qwen3.8-flash; Singapore / International. Direct text-only standard rate; flat across 0<input≤1M pricing range. Implicit and explicit cache reads are both USD 0.016/M, but explicit cache creation costs USD 0.2/M; do not assume generic 20%/10%/125% cache ratios. Thinking output includes reasoning tokens. Maximum input 991,808 non-thinking / 983,616 thinking. International text batch unsupported.
  • Supplier API identity: qwen3.8-flash.
  • Deployment scope: International.
  • Source checked: 2026-11-02.
AlibabaQwen3.8 Flash (Global)Budget · Global scopeVerify provider price ↗$0.113$0.014$0.3821,000,000 tokens
  • Calculator id is scope-qualified; API model=qwen3.8-flash. Global at Frankfurt/Virginia/Tokyo/Hong Kong, with mainland and overseas inference. Flat standard text rate. Both implicit and explicit cache reads USD 0.014/M; explicit cache creation USD 0.177/M. These are regional exceptions to generic cache ratios. Maximum input 991,808 / thinking 983,616. Batch unsupported in Global regions. US-only, EU-only and Hong Kong-only scopes separately list standard USD 0.15 input / 0.47 output / 0.016 implicit cache per M; no single multiplier is assumed.
  • Supplier API identity: qwen3.8-flash.
  • Deployment scope: Global.
  • Source checked: 2026-11-02.
AlibabaQwen3.7 Flash (International)BudgetVerify provider price ↗$0.03$0.006$0.131,000,000 tokens
  • API model=qwen3.7-flash, currently equivalent to qwen3.7-flash-2026-07-15; International / Singapore. Distinct currently recommended lower-cost tiered Flash model, not relabeled as 3.8. All three total-input tiers are in contextRates; choose the greatest satisfied threshold, not one long-context rate for every prompt. Explicit creation/read USD 0.038/0.003 per M at input≤32K, 0.125/0.01 at 32K<input≤256K, 0.25/0.02 at 256K<input≤1M. Implicit cache values are in rate. Maximum input 991,808 / thinking 983,616. Thinking tokens billed as output.
  • Supplier API identity: qwen3.7-flash.
  • Deployment scope: International.
  • Source checked: 2026-11-02.
  • Data-use warning: Pricing summary has a 50% batch badge, but the model-specific Singapore capability table says batch unsupported and current Batch File scope omits qwen3.7-flash. Do not apply a Singapore batch discount.
  • Input tier above 0 through 32,000 tokens: Input $0.03/M; cached input $0.006/M; output $0.13/M.
  • Input tier above 32,000 through 256,000 tokens: Input $0.10/M; cached input $0.02/M; output $0.40/M.
  • Input tier above 256,000 through 1,000,000 tokens: Input $0.20/M; cached input $0.04/M; output $0.80/M.
AlibabaQwen3.7 Flash (Global)Budget · Global scopeVerify provider price ↗$0.028$0.006$0.111,000,000 tokens
  • API model=qwen3.7-flash; scope-qualified calculator id. Global at Frankfurt/Virginia/Tokyo/Hong Kong; inference may include Chinese mainland. All three total-input tiers preserved. Explicit creation/read USD 0.034/0.003 per M at input≤32K, 0.103/0.008 at 32K<input≤256K, 0.206/0.017 above 256K. Implicit cache decimals are exact published values, not a generic-ratio calculation. Maximum input 991,808 / thinking 983,616; thinking tokens billed as output. Global batch unsupported.
  • Supplier API identity: qwen3.7-flash.
  • Deployment scope: Global.
  • Source checked: 2026-11-02.
  • Input tier above 0 through 32,000 tokens: Input $0.028/M; cached input $0.006/M; output $0.11/M.
  • Input tier above 32,000 through 256,000 tokens: Input $0.083/M; cached input $0.017/M; output $0.33/M.
  • Input tier above 256,000 through 1,000,000 tokens: Input $0.165/M; cached input $0.033/M; output $0.66/M.
AlibabaQwen3 Coder Flash (International)Legacy · Coding specialistVerify provider price ↗$0.30$0.06$1.501,000,000 tokens
  • API model=qwen3-coder-flash, equivalent to qwen3-coder-flash-2025-07-28. Explicitly legacy/no longer recommended in current text guide, but remains directly documented and priced; included as the currently documented dedicated Coder family, not as baseline preservation. Singapore / International. Four total-input tiers apply to all tokens. Explicit creation/read USD 0.375/0.03, 0.625/0.05, 1/0.08, 2/0.16 per M across ≤32K, >32K–128K, >128K–256K, >256K–1M respectively. Implicit cache in rate; maximum input 997,952; maximum output 65,536. Batch unsupported. Do not use this row for retired Coder Plus or Coder Next.
  • Supplier API identity: qwen3-coder-flash.
  • Deployment scope: International.
  • Source checked: 2026-11-02.
  • Input tier above 0 through 32,000 tokens: Input $0.30/M; cached input $0.06/M; output $1.50/M.
  • Input tier above 32,000 through 128,000 tokens: Input $0.50/M; cached input $0.10/M; output $2.50/M.
  • Input tier above 128,000 through 256,000 tokens: Input $0.80/M; cached input $0.16/M; output $4.00/M.
  • Input tier above 256,000 through 1,000,000 tokens: Input $1.60/M; cached input $0.32/M; output $9.60/M.
AlibabaQwen3 Coder Flash (Global)Legacy · Coding / GlobalVerify provider price ↗$0.144$0.029$0.5741,000,000 tokens
  • API model=qwen3-coder-flash; calculator id is scope-qualified. Legacy but currently documented dedicated coding model, Global at Frankfurt or Virginia. Inference may include Chinese mainland. Four total-input tiers retained. Explicit creation/read USD 0.18/0.014, 0.27/0.022, 0.449/0.036, 0.896/0.072 per M across ≤32K, >32K–128K, >128K–256K, >256K–1M respectively. Implicit cache uses exact published rounded decimals. Maximum input 997,952 and output 65,536. Batch unsupported.
  • Supplier API identity: qwen3-coder-flash.
  • Deployment scope: Global.
  • Source checked: 2026-11-02.
  • Input tier above 0 through 32,000 tokens: Input $0.144/M; cached input $0.029/M; output $0.574/M.
  • Input tier above 32,000 through 128,000 tokens: Input $0.216/M; cached input $0.044/M; output $0.861/M.
  • Input tier above 128,000 through 256,000 tokens: Input $0.359/M; cached input $0.072/M; output $1.434/M.
  • Input tier above 256,000 through 1,000,000 tokens: Input $0.717/M; cached input $0.144/M; output $3.584/M.
AlibabaQwen3.8 2.4T A95B (International)Frontier · Open-weight flagshipVerify provider price ↗$2.00$0.25$6.001,000,000 tokens
  • Exact API model=qwen3.8-2.4t-a95b, a distinct open-source flagship API identity; not relabeled as qwen3.8-max. Singapore / International only for the included scope; no documented Global USD table, so no Global row inferred from Beijing. Standard text input/output, both thinking and non-thinking. Implicit cache USD 0.25/M; explicit creation/read USD 2.5/0.17 per M. Maximum input 991,808 / thinking 983,616. Batch unsupported.
  • Supplier API identity: qwen3.8-2.4t-a95b.
  • Deployment scope: International.
  • Source checked: 2026-11-02.
AlibabaQwen3.8 27B (International)Balanced · Open-weight denseVerify provider price ↗$0.50$0.10$3.001,000,000 tokens
  • Exact API model=qwen3.8-27b; current distinct dense-model tier. Singapore / International. Standard text input/output, both thinking and non-thinking. Implicit cache in rate; explicit creation/read USD 0.625/0.05 per M. No documented Global price table, so no Global price inferred from Beijing. Maximum input 991,808 / thinking 983,616; batch unsupported.
  • Supplier API identity: qwen3.8-27b.
  • Deployment scope: International.
  • Source checked: 2026-11-02.
AlibabaQwen3.6 35B A3B (International)Budget · Open-weight small MoEVerify provider price ↗$0.375Not listed$2.25262,144 tokens
  • API model=qwen3.6-35b-a3b; distinct currently documented small sparse MoE tier, with no newer 35B-A3B family identity listed. Singapore / International. Both thinking/non-thinking text output prices match; reasoning tokens billed as output. Cache unsupported, so cachedInputPerM omitted, not zero. Batch unsupported. Exact context 262,144; maximum input 260,096 / thinking 258,048; output 65,536. Summary pricing uses 0<Token≤256K, where K=1000; model page instead gives exact input capacity. No higher prompt-price tier documented; do not invent one above 256,000.
  • Supplier API identity: qwen3.6-35b-a3b.
  • Deployment scope: International.
  • Source checked: 2026-11-02.
AlibabaQwen3.6 35B A3B (Global)Budget · Small MoE / GlobalVerify provider price ↗$0.248Not listed$1.485262,144 tokens
  • API model=qwen3.6-35b-a3b; scope-qualified calculator id. Global at Frankfurt or Virginia, with possible mainland inference. Distinct latest documented small MoE tier. Both thinking/non-thinking output prices match. Cache and batch unsupported; omit cache rate rather than zero. Exact context 262,144; input 260,096 / thinking 258,048. Summary pricing range is 0<Token≤256K (decimal); exact model page capacity exceeds this slightly and gives a flat original price. No undocumented higher tier inferred.
  • Supplier API identity: qwen3.6-35b-a3b.
  • Deployment scope: Global.
  • Source checked: 2026-11-02.
CohereCommand AFrontierVerify provider price ↗$2.50Not listed$10.00256,000 tokens
  • Direct Cohere production Chat API, standard text input/output, published dollar prices per 1M tokens. Maximum output 8,000 tokens. Distinct from Command A+ and Command A Reasoning. No regional, batch or long-context price adjustment was documented in the checked standard pricing sources.
  • Supplier API identity: command-a-03-2025.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:43Z.
  • Data-use warning: No official cached-input rate or cache-write, TTL, storage or eligibility schedule was located for this direct API model; cachedInputPerM is omitted, not zero. Tier is a calculator classification, not a supplier billing tier.
CohereCommand R (08-2024)BalancedVerify provider price ↗$0.15Not listed$0.60128,000 tokens
  • Direct Cohere production Chat API, standard text input/output. Maximum output 4,000 tokens. The exact API identity is command-r-08-2024, not the deprecated command-r alias, which points to March 2024. Supplier recommends Command A for most use cases but still lists this version as Live.
  • Supplier API identity: command-r-08-2024.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:43Z.
  • Data-use warning: No official cached-input rate, cache-write fee, TTL, storage fee, regional surcharge or automatic long-context billing trigger was located. Optional price fields are omitted rather than inferred. Tier is a calculator classification.
CohereCommand R7B (12-2024)BudgetVerify provider price ↗$0.0375Not listed$0.15128,000 tokens
  • Direct Cohere production Chat API, standard text input/output. Exact published input rate is $0.0375 per 1M, not a rounded $0.04. Maximum output 4,000 tokens. Documentation explicitly supports chatbots and code assistants.
  • Supplier API identity: command-r7b-12-2024.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:43Z.
  • Data-use warning: No official cached-input rate or cache-write/TTL/storage schedule was located; cachedInputPerM is omitted. No batch or automatic long-context tier was documented. Tier is a calculator classification.
CohereCommand R+ (08-2024)Legacy · Live older generationVerify provider price ↗$2.50Not listed$10.00128,000 tokens
  • Distinct live August 2024 API model, not the deprecated command-r-plus alias for April 2024. Direct standard text Chat API; maximum output 4,000. The Legacy calculator tier reflects an older generation and the supplier recommendation to use Command A, not API retirement.
  • Supplier API identity: command-r-plus-08-2024.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:19:43Z.
  • Data-use warning: Pricing-page FAQ groups this version under existing-customer legacy pricing, while its dedicated model page has a public pricing card and the API catalog marks it Live. Availability is based on the dedicated page and catalog; no paid call was made. No official cache rates or cache-write/TTL/storage schedule were located.
CohereAya Expanse 32BBalancedVerify provider price ↗$0.50Not listed$1.50128,000 tokens
  • Included as a distinct publicly priced live general multilingual text/chat model. Direct API billing, not the open-weight license or a trial rate. Maximum output 4,000 tokens. Research-family designation does not make the documented paid API rate a free trial offer.
  • Supplier API identity: c4ai-aya-expanse-32b.
  • Deployment scope: Supplier default.
  • Source checked: 2026-11-02T22:20:42Z.
  • Data-use warning: The pricing FAQ mentions both 8B and 32B, but the 8B identity retired on April 4, 2026 and is excluded. No official cached-input, cache-write/TTL/storage, regional or batch schedule was located. Tier is a calculator classification.

Cache, batch, fast mode and US inference apply only when the selected row documents the relevant rate or multiplier. Provider-specific conditions are shown directly from the catalog above; unsupported rows keep their applicable rate.

If the prompt is client or regulated data, a cheaper token tier is not a governance decision. Review AI governance before you send it.

The standard list-price table keeps documented peak rates for direct DeepSeek. Optional off-peak pricing changes calculator estimates, cards, comparison and bulk results, not this list-price table.

Your text

About four characters per token. Planning estimate, not a provider tokenizer.

Published input and output rate for the selected row

List price dated .

Optional switches

Only if that row documents them: cache-hit share, batch rate, fast mode or US-only inference. Check the row's published conditions before applying a setting.

Manus credits

Not on this dollar scale. No per-million token rate on this page.

How this calculator prices a call

Diagram of the token calculator method. Text is estimated at about four characters per token, then priced with the selected model's published input and output rates as of November 2, 2026. Cache, batch, fast mode and US inference apply only where this page says the provider documents them. Manus credits sit on a separate branch with no token price.

Try Manus free - Run multi-step AI workflows (research, content, analysis) with bonus credits. No credit card required.

Manus uses credits for agent tasks; it is not presented as a normal dollars-per-million-token API model here.

Claim free Manus credits →

Calculator settings

Prices as of November 2, 2026 - verify on provider pages before committing spend.

Your prompt or input text

0

Characters

0

Words

0

Estimated tokens

Estimate only: the page uses about four characters per token for English planning. Tokenization varies by provider, language, code and model; use the provider tokenizer for exact deployment counts.

Verified model estimates

Last updated: (page amendment) · source links are shown on every row.

Filters change the catalog, not your saved model selection. Sorting also applies to Compare and Bulk. These browsing preferences stay on this page, not in shared comparison links.

Tier

USD per million tokens. Leave either maximum blank for no limit; a maximum includes matching prices.

Sorting and price limits use the currently applied per-million uncached input/output rates and pricing settings. Cache mix affects total-cost ordering only, not these unit-rate filters.

Showing 93 of 93 models.

6 of 6 selected rows list a cache-hit rate in this mode

Only direct DeepSeek rows with a documented schedule are affected, not Mistral-hosted DeepSeek or an arbitrary discount. UTC peak: Mon–Fri 01:00–04:00 & 06:00–10:00, excluding Chinese public holidays. All other hours, weekends and Chinese public holidays are off-peak. Choose manually for planning: this tool does not detect the current hour or holidays. Verify provider timing terms for your requests.

Settings apply only where the row documents them; unsupported rows retain their applicable rate. Cache hits do not include separate cache-write or storage charges. Batch availability can differ for long-context requests. Fast mode takes precedence over batch on supported rows; verify provider compatibility and regional terms.

Mistral

Ministral 3 3B

GA

Input

$0.10/M

Cached

$0.01/M

Output

$0.10/M

$0.000050 per call

standard rate

Distinct 3B tier of Ministral 3. Direct Mistral global/default Standard Tier text rates in USD per 1 million tokens. Card displays 256k context; exact integer not established. This is not retired Ministral 3B ministral-3b-2410. See sourceNotes for cache conditions.

Verify provider price ↗

Alibaba

Qwen3.7 Flash (Global)

Global scope

Input

$0.028/M

Cached

$0.006/M

Output

$0.11/M

$0.000055 per call

standard rate

500 request1.00M context

API model=qwen3.7-flash; scope-qualified calculator id. Global at Frankfurt/Virginia/Tokyo/Hong Kong; inference may include Chinese mainland. All three total-input tiers preserved. Explicit creation/read USD 0.034/0.003 per M at input≤32K, 0.103/0.008 at 32K<input≤256K, 0.206/0.017 above 256K. Implicit cache decimals are exact published values, not a generic-ratio calculation. Maximum input 991,808 / thinking 983,616; thinking tokens billed as output. Global batch unsupported.

Verify provider price ↗

Alibaba

Qwen3.7 Flash (International)

Input

$0.03/M

Cached

$0.006/M

Output

$0.13/M

$0.000065 per call

standard rate

500 request1.00M context

Data-use warning: Pricing summary has a 50% batch badge, but the model-specific Singapore capability table says batch unsupported and current Batch File scope omits qwen3.7-flash. Do not apply a Singapore batch discount.

API model=qwen3.7-flash, currently equivalent to qwen3.7-flash-2026-07-15; International / Singapore. Distinct currently recommended lower-cost tiered Flash model, not relabeled as 3.8. All three total-input tiers are in contextRates; choose the greatest satisfied threshold, not one long-context rate for every prompt. Explicit creation/read USD 0.038/0.003 per M at input≤32K, 0.125/0.01 at 32K<input≤256K, 0.25/0.02 at 256K<input≤1M. Implicit cache values are in rate. Maximum input 991,808 / thinking 983,616. Thinking tokens billed as output.

Verify provider price ↗

Mistral

Ministral 3 8B

GA

Input

$0.15/M

Cached

$0.015/M

Output

$0.15/M

$0.000075 per call

standard rate

Distinct 8B tier of Ministral 3. Direct Mistral global/default Standard Tier text rates in USD per 1 million tokens. Card displays 256k context; exact integer not established. This is not retired Ministral 8B ministral-8b-2410. See sourceNotes for cache conditions.

Verify provider price ↗

Cohere

Command R7B (12-2024)

Input

$0.0375/M

Cached

-

Output

$0.15/M

$0.000075 per call

standard rate

500 request128.0K context

Data-use warning: No official cached-input rate or cache-write/TTL/storage schedule was located; cachedInputPerM is omitted. No batch or automatic long-context tier was documented. Tier is a calculator classification.

Direct Cohere production Chat API, standard text input/output. Exact published input rate is $0.0375 per 1M, not a rounded $0.04. Maximum output 4,000 tokens. Documentation explicitly supports chatbots and code assistants.

Verify provider price ↗

Meta Muse

Muse Spark 1.3 Contributor

Contributor

Input

$0.10/M

Cached

$0.002/M

Output

$0.20/M

$0.000100 per call

standard rate

500 request1.05M context

Data-use warning: Contributor pricing permits Meta to use prompts and completions to train future Meta models. This also covers submitted code, files and context. Do not submit sensitive, confidential or personal information, including confidential software code. Previously Standard content reused in a Contributor request becomes Contributor content. Contributor variants are unavailable to ZDR teams and offer no per-traffic training opt-out.

Exact API identity: muse-spark-1.3-contributor; retained calculator ID is not the API ID. Public direct pay-as-you-go Contributor mode; official supplier conditional rates, not an assumed reseller discount. No percentage discount or expiry supplied in baseline; current docs state absolute rates without a dated promotion expiry. Direct default non-batch text USD per 1M tokens, no long-context premium. Same Spark 1.3 model but max reasoning is unavailable in Contributor mode. Automatic cache-hit pricing only; uncached tokens billed at input rate. Separate cache-write/storage rates not published; default in-memory TTL not fixed, optional Responses 24h is a hint.

Verify provider price ↗

Meta Muse

Muse Spark 1.2 Contributor

Contributor

Input

$0.10/M

Cached

$0.002/M

Output

$0.20/M

$0.000100 per call

standard rate

500 request1.05M context

Data-use warning: Contributor pricing permits Meta to use prompts and completions to train future Meta models. This also covers submitted code, files and context. Do not submit sensitive, confidential or personal information, including confidential software code. Previously Standard content reused in a Contributor request becomes Contributor content. Contributor variants are unavailable to ZDR teams and offer no per-traffic training opt-out.

Exact API identity: muse-spark-1.2-contributor; retained calculator ID is not the API ID. Previous Spark version remains documented and publicly priced; Legacy denotes previous generation, not retired. Official supplier Contributor conditional rates, not an assumed reseller discount. No percentage offer or expiry supplied; current absolute rates have no documented promotion expiry. Direct default non-batch text USD per 1M tokens; no long-context premium. Cache-hit pricing only, misses billed at input price. Separate cache-write/storage prices not published; default in-memory cache has no fixed TTL, optional Responses 24h is a hint.

Verify provider price ↗

Mistral

Ministral 3 14B

GA

Input

$0.20/M

Cached

$0.02/M

Output

$0.20/M

$0.000100 per call

standard rate

Distinct 14B tier of Ministral 3. Direct Mistral global/default Standard Tier text rates in USD per 1 million tokens; not self-hosting cost. Card displays 256k context; exact integer not established. See sourceNotes for cache conditions.

Verify provider price ↗

Alibaba

Qwen3.8 Flash (Global)

Global scope

Input

$0.113/M

Cached

$0.014/M

Output

$0.382/M

$0.000191 per call

standard rate

500 request1.00M context

Calculator id is scope-qualified; API model=qwen3.8-flash. Global at Frankfurt/Virginia/Tokyo/Hong Kong, with mainland and overseas inference. Flat standard text rate. Both implicit and explicit cache reads USD 0.014/M; explicit cache creation USD 0.177/M. These are regional exceptions to generic cache ratios. Maximum input 991,808 / thinking 983,616. Batch unsupported in Global regions. US-only, EU-only and Hong Kong-only scopes separately list standard USD 0.15 input / 0.47 output / 0.016 implicit cache per M; no single multiplier is assumed.

Verify provider price ↗

Google

Gemini 2.5 Flash-Lite

Legacy; existing users only

Input

$0.10/M

Cached

$0.01/M

Output

$0.40/M

$0.000200 per call

standard rate

500 request1.05M context

Data-use warning: Not generally available to new API users/projects; cannot be used as an unrestricted current-model recommendation. Cache-hit price excludes storage and does not establish explicit creation/write pricing. Exact model minimum is absent from inspected developer cache table; not guessed. Explicit caching Beta through GenerateContent only; default TTL one hour. Output includes thinking, and tools/grounding cost extra.

Exact stable API identity gemini-2.5-flash-lite; documented baseline legacy retained, not added as a newest/general-availability recommendation. Google limits 2.5 access to users who actively used it in the past. This model is explicitly not deprecated and continues through the API until further notice. Standard and Batch explicit storage add $1.00 per 1M cached tokens per hour; Batch cache-hit rate remains $0.01, not automatically halved. Prices are text scope, excluding separate audio input rates. No prompt-length price tier is documented.

Verify provider price ↗

Alibaba

Qwen3.8 Flash (International)

Thinking + coding

Input

$0.15/M

Cached

$0.016/M

Output

$0.47/M

$0.000235 per call

standard rate

500 request1.00M context

API model=qwen3.8-flash; Singapore / International. Direct text-only standard rate; flat across 0<input≤1M pricing range. Implicit and explicit cache reads are both USD 0.016/M, but explicit cache creation costs USD 0.2/M; do not assume generic 20%/10%/125% cache ratios. Thinking output includes reasoning tokens. Maximum input 991,808 non-thinking / 983,616 thinking. International text batch unsupported.

Verify provider price ↗

OpenAI

GPT-6 Luna

High volume

Input

$0.10/M

Cached

$0.01/M

Output

$0.50/M

$0.000250 per call

standard rate

500 request1.05M context

Data-use warning: The supplied 15% reseller discount is unverified and is not applied. Cache-hit price does not include cache writes.

Requested identity and baseline retained. Public text API; Responses and Chat Completions supported. Maximum input 922,000/output 128,000. Current model page now explicitly documents >272K input-token long-context trigger, resolving the baseline's missing-threshold warning. Cache writes USD 0.125/1M for 30m or 0.25/1M for 12h; long-context writes 0.25/0.50. Long rates apply to the full request. Flex shares Batch pricing. Fast long-context input/output/cached 0.40/1.50/0.04 per 1M. EU residency supports Standard, Fast, Flex and Batch. Chat Completions function calling requires reasoning_effort none; use Responses for built-in tools.

Verify provider price ↗

Anthropic

Claude Haiku 5.5

Budget

Input

$0.10/M

Cached

$0.01/M

Output

$0.50/M

$0.000250 per call

standard rate

500 request1.00M context

Data-use warning: The supplied $0.10/$0.50/$0.01 list rates are verified only for prompts of at most 100,000 tokens, not the full 1M context. Cache write costs are separate. Supplied reseller discount is not verified.

Added requested/latest Haiku model with exact public Claude API identity. Standard global/default synchronous text rates in USD per million tokens. Base rates apply to prompts up to and including 100,000 tokens. A request with more than 100,000 total prompt/input tokens uses the higher input, output, cache-read and cache-write rates for that entire request, not just excess tokens. Prompt length includes uncached input, cache reads and cache writes; output length is not the trigger. Each request is assessed independently, and earlier requests are not repriced. Up to 100k: 5m writes $0.125/MTok; 1h writes $0.20/MTok. Over 100k: 5m writes $0.625/MTok; 1h writes $1/MTok. Cacheable minimum prefix 512 tokens. Batch table documents both tiers; cache hits follow the documented 50% all-usage discount. US-only inference applies 1.1x to every category in both prompt-length tiers.

Verify provider price ↗

Z.AI

GLM-5.3-Flash

Requested model verified

Input

$0.15/M

Cached

$0.03/M

Output

$0.50/M

$0.000250 per call

standard rate

500 request1.00M context

Data-use warning: Do not apply an assumed 50% discount. Supplier blog's '$0.045 per task (discounted)' benchmark is not a per-token rate or current promotion contract. Coding Plan off-peak 50% points usage is not standard API token pricing.

Exact requested model, not GLM-5.3 or GLM-5.3-FlashX. A native multimodal model with an explicitly supported plain-text API mode and text output; rates here cover text only, not image/video/file processing. Docs publish 1M context and mandatory reasoning. Standard prices are retained; supplied 0.075/0.25/0.015 discounted rates are not verified as a currently eligible direct API offer. Implicit cache new content uses standard input billing; cached-input price is for hits, not a full cache-write/storage price. Storage is currently described as 'Limited-time Free', without an end date or numerical TTL. No documented default long-context price trigger.

Verify provider price ↗

Mistral

Z.ai GLM 5.3 Flash (Mistral-hosted)

GA · Third-party

Input

$0.15/M

Cached

$0.03/M

Output

$0.50/M

$0.000250 per call

standard rate

Data-use warning: Model-specific published cached input is $0.03/M, not the blanket 10%-of-input statement in Mistral's caching guide. Use the explicit current model card/table; no inferred discount. Cache-write, storage and TTL charges are not separately specified in inspected documentation.

Distinct Flash tier, Z.ai model hosted without Mistral modifications. Mistral direct global/default Standard Tier text rates in USD per 1 million tokens, not Z.ai API rates. Card displays 1M context and 128k max output; exact integer limits not established.

Verify provider price ↗

Tencent

Hunyuan Hy3

Input

$0.132/M

Cached

$0.033/M

Output

$0.528/M

$0.000264 per call

standard rate

Data-use warning: Cache price is actual cached-input consumption only. Both 5-minute and 1-hour cache-write cells are '-' rather than numeric zero. Reviewed HY docs do not publish an exact cache TTL, storage price, minimum eligible prompt length or separately quantified write charge; do not assume free cache writes/storage or guaranteed hits. Account/model activation and the billing guide's USD 10 frozen amount apply; free-trial and token-plan credits are not standard token-rate discounts.

Current official-version API identity is hy3, not hy3-preview. Tencent describes general-purpose, coding, long-text, reasoning and agent capabilities. Scope is Tencent-operated TokenHub pay-as-you-go standard online non-batch text requests at the Singapore/Global endpoint. Model documentation renders context as 256k, maximum input as 192k and maximum output as 128k; exact integer contextTokens is omitted rather than inferred from abbreviations. Supports Chat Completions, Responses and Anthropic Messages after activation. No length-tier or peak/off-peak condition is listed; no batch/fast rate is asserted.

Verify provider price ↗

Alibaba

Qwen3 Coder Flash (Global)

Coding / Global

Input

$0.144/M

Cached

$0.029/M

Output

$0.574/M

$0.000287 per call

standard rate

500 request1.00M context

API model=qwen3-coder-flash; calculator id is scope-qualified. Legacy but currently documented dedicated coding model, Global at Frankfurt or Virginia. Inference may include Chinese mainland. Four total-input tiers retained. Explicit creation/read USD 0.18/0.014, 0.27/0.022, 0.449/0.036, 0.896/0.072 per M across ≤32K, >32K–128K, >128K–256K, >256K–1M respectively. Implicit cache uses exact published rounded decimals. Maximum input 997,952 and output 65,536. Batch unsupported.

Verify provider price ↗

OpenAI

GPT-4o mini

Deprecated

Input

$0.15/M

Cached

$0.075/M

Output

$0.60/M

$0.000300 per call

standard rate

500 request128.0K context

Data-use warning: Deprecated; scheduled removal from API January 14, 2027. Recheck/remove before shutdown.

Published baseline legacy retained with currently documented Standard rates. Stable alias maps to gpt-4o-mini-2024-07-18; maximum output 16,384. Responses/Chat Completions/Batch supported. Earlier implicit cache: no extra write charge; in_memory generally 5–10 minutes inactive, up to one hour. No published Batch cached rate ('-'), so omitted rather than assumed zero.

Verify provider price ↗

Mistral

Mistral Small 4

GA

Input

$0.15/M

Cached

$0.015/M

Output

$0.60/M

$0.000300 per call

standard rate

General text, reasoning and coding in a single model. Direct Mistral global/default Standard Tier USD per 1 million text tokens. Model card displays 256k context; exact integer not established, so contextTokens is omitted. Cache eligibility and undisclosed write/TTL/storage details are in sourceNotes.

Verify provider price ↗

Cohere

Command R (08-2024)

Input

$0.15/M

Cached

-

Output

$0.60/M

$0.000300 per call

standard rate

500 request128.0K context

Data-use warning: No official cached-input rate, cache-write fee, TTL, storage fee, regional surcharge or automatic long-context billing trigger was located. Optional price fields are omitted rather than inferred. Tier is a calculator classification.

Direct Cohere production Chat API, standard text input/output. Maximum output 4,000 tokens. The exact API identity is command-r-08-2024, not the deprecated command-r alias, which points to March 2024. Supplier recommends Command A for most use cases but still lists this version as Live.

Verify provider price ↗

Mistral

Codestral 25.08

GA

Input

$0.30/M

Cached

$0.03/M

Output

$0.90/M

$0.000450 per call

standard rate

Current Codestral code-generation/completion API, not Codestral Embed or retired Codestral 25.01. Direct Mistral global/default Standard Tier USD per 1 million text tokens. Card displays 128k context; exact integer not established. See sourceNotes for shared-prefix/FIM cache conditions.

Verify provider price ↗

Alibaba

Qwen3.7 Plus (Global)

Global scope

Input

$0.276/M

Cached

$0.056/M

Output

$1.101/M

$0.000550 per call

standard rate

500 request1.00M context

Data-use warning: Official Global list-price table labels limited-time night 60% off / daytime 20% off, with night 22:00–08:00 UTC+8 based on billing time. Expiry and full eligibility are not established; no discounts applied.

API model=qwen3.7-plus (calculator id is scope-qualified); Global at Frankfurt/Virginia/Tokyo/Hong Kong, including mainland inference resources. Total-input tier applies to all tokens. Both thinking/non-thinking output modes use the listed output rate. Use published rounded cache prices rather than multiplying the general cache ratio. Explicit creation/read USD 0.344/0.028 per M at input≤256K and 1.032/0.083 per M above 256K. Maximum input 991,808 / thinking 983,616. Global batch unsupported. US-only scope separately lists standard USD 0.4/1.6/0.08 per M at input≤256K and 1.2/4.8/0.24 above; not represented by an assumed multiplier.

Verify provider price ↗

OpenAI

GPT-5.6 Luna

Input

$0.20/M

Cached

$0.02/M

Output

$1.20/M

$0.000600 per call

standard rate

500 request1.05M context

Active, separately priced previous Luna tier. Maximum input 922,000/output 128,000. Cache writes USD 0.25/1M for 30m or 0.50/1M for 12h; long writes 0.50/1. Long rates cover full request above 272,000 input tokens. Flex shares Batch rates. Fast long input/output/cached 0.80/3.60/0.08 per 1M.

Verify provider price ↗

DeepSeek

DeepSeek V4.1 Flash

Peak/off-peak pricing

Input

$0.30/M

Cached

$0.006/M

Output

$1.20/M

$0.000600 per call

standard rate

Data-use warning: Do not display the peak rate as an all-hours flat price or the off-peak rate as an unconditional list price. Preserve both time schedules and Chinese-holiday eligibility. No supplier confirmation was found for the supplied additional discounted rates. Cache-hit rate alone must not be interpreted as a premium cache-write price or guaranteed cache availability.

Current stable public API identity is deepseek-flash, served by DeepSeek-V4.1-Flash-1017. Direct standard non-batch text rates in USD per 1M tokens apply to both thinking and non-thinking modes; thinking is enabled by default. rate contains the published PEAK schedule as a conservative untimed calculator baseline, not an all-hours flat price. Official OFF-PEAK rates are input/cache-miss $0.15, output $0.60, and cached-input/cache-hit $0.003 per 1M tokens. Peak hours are 01:00-04:00 and 06:00-10:00 UTC Monday-Friday, excluding Chinese public holidays; every other hour, all weekends and Chinese public holidays are off-peak. These schedules apply to API users generally and are not a reseller promotion. The documentation states 1M total context and a maximum output of 384K (393216 tokens); contextTokens is omitted because the checked public pages do not give an exact integer for the 1M context limit. Automatic disk-prefix caching is enabled for all users. Cache-hit pricing applies only to fully matching persisted prefix units; cache misses, including initial uncached requests, use the input rate. The current cache guide describes persistence at user-input/model-output boundaries, detected shared prefixes and fixed intervals, best-effort hits, construction taking seconds, and unused-cache eviction usually within hours to days rather than a guaranteed TTL. The historical official cache announcement says there is no additional cache-service fee and storage is free; the current pricing page publishes no separate cache-write or storage charge. Its historical token prices and 64-token storage rule are not used as current pricing or a current minimum. No documented long-context token-price uplift, batch schedule, fast surcharge, or US-residency multiplier was found in the current pricing page; those fields are omitted. This row covers text only, not image input, search/fetch tools, or storage products.

Verify provider price ↗

Mistral

DeepSeek V4.1 Flash 1017 (Mistral-hosted)

Public Preview · Third-party

Input

$0.30/M

Cached

$0.007/M

Output

$1.20/M

$0.000600 per call

standard rate

Data-use warning: Public Preview permits silent updates. Explicit cached input $0.007/M differs from the generic caching guide's 10%-of-input statement; exact current model card/table wins. No assumed cache-write/storage rate or TTL is added.

DeepSeek model hosted by Mistral without modifications. Exact card/API identity contains v4-1 although documentation URL contains v4; not relabeled as another DeepSeek model. One currently offered identity is retained, not a snapshot catalog. Direct Mistral global/default Standard Tier text-token USD rates per 1 million tokens. Card displays 1M context and 128k max output; exact integer limits not established.

Verify provider price ↗

Z.AI

GLM-5.3-FlashX

High-speed tier

Input

$0.37/M

Cached

$0.075/M

Output

$1.25/M

$0.000625 per call

standard rate

500 request1.00M context

Distinct named public API tier, not a guessed speed multiplier on Flash. Supplier says now live with inference speed of 200 tokens/s; this is a performance claim, not a guarantee. Plain-text requests and text output are supported; only text rates are captured. Published context 1M. New cache content uses normal input billing; cache-hit rate does not price all writes/storage. Storage Limited-time Free with no documented expiry or numerical TTL. No default long-context surcharge documented. Represented as its own API model, not duplicated as fastRate on Flash.

Verify provider price ↗

Z.AI

GLM-5.3-FlashX-Legal

Input

$0.37/M

Cached

$0.075/M

Output

$1.25/M

$0.000625 per call

standard rate

500 request1.00M context

Supplier-managed, separately named and priced reasoning/text API model optimized for legal research, contract review and drafting; not a customer fine-tuning service or enterprise quote. Included as a distinct active GLM-5.3 family identity. Optional external legal MCP integrations are not included in these token rates. Plain-text scope only. Published context 1M. Cache new content uses standard input billing; hits use listed cached-input rate; storage is Limited-time Free with unspecified expiry and numerical TTL.

Verify provider price ↗

Alibaba

Qwen3.6 35B A3B (Global)

Small MoE / Global

Input

$0.248/M

Cached

-

Output

$1.485/M

$0.000743 per call

standard rate

500 request262.1K context

API model=qwen3.6-35b-a3b; scope-qualified calculator id. Global at Frankfurt or Virginia, with possible mainland inference. Distinct latest documented small MoE tier. Both thinking/non-thinking output prices match. Cache and batch unsupported; omit cache rate rather than zero. Exact context 262,144; input 260,096 / thinking 258,048. Summary pricing range is 0<Token≤256K (decimal); exact model page capacity exceeds this slightly and gives a flat original price. No undocumented higher tier inferred.

Verify provider price ↗

Google

Gemini 3.1 Flash-Lite

Budget

Input

$0.25/M

Cached

$0.025/M

Output

$1.50/M

$0.000750 per call

standard rate

500 request1.05M context

Data-use warning: Output includes thinking. Cache-hit rate excludes storage and does not establish explicit cache-write cost. Implicit caching applies to Gemini 2.5 and newer but exact model minimum is not in the inspected developer cache table; not guessed. Explicit caching Beta on GenerateContent only, default TTL one hour. Additional non-cached tokens, tools, and grounding charges are excluded from listed token rates.

Exact stable API identity gemini-3.1-flash-lite; baseline calculator ID preserved. These are text input and text output prices, not the separate audio input prices. Standard explicit-cache storage adds $1.00 per 1M tokens per hour; Batch table explicitly lists $0.50 per 1M tokens per hour. No prompt-length pricing tier or promo expiry documented. Earliest published shutdown May 7, 2027; recommended replacement gemini-3.5-flash-lite.

Verify provider price ↗

Alibaba

Qwen3 Coder Flash (International)

Coding specialist

Input

$0.30/M

Cached

$0.06/M

Output

$1.50/M

$0.000750 per call

standard rate

500 request1.00M context

API model=qwen3-coder-flash, equivalent to qwen3-coder-flash-2025-07-28. Explicitly legacy/no longer recommended in current text guide, but remains directly documented and priced; included as the currently documented dedicated Coder family, not as baseline preservation. Singapore / International. Four total-input tiers apply to all tokens. Explicit creation/read USD 0.375/0.03, 0.625/0.05, 1/0.08, 2/0.16 per M across ≤32K, >32K–128K, >128K–256K, >256K–1M respectively. Implicit cache in rate; maximum input 997,952; maximum output 65,536. Batch unsupported. Do not use this row for retired Coder Plus or Coder Next.

Verify provider price ↗

Cohere

Aya Expanse 32B

Input

$0.50/M

Cached

-

Output

$1.50/M

$0.000750 per call

standard rate

500 request128.0K context

Data-use warning: The pricing FAQ mentions both 8B and 32B, but the 8B identity retired on April 4, 2026 and is excluded. No official cached-input, cache-write/TTL/storage, regional or batch schedule was located. Tier is a calculator classification.

Included as a distinct publicly priced live general multilingual text/chat model. Direct API billing, not the open-weight license or a trial rate. Maximum output 4,000 tokens. Research-family designation does not make the documented paid API rate a free trial offer.

Verify provider price ↗

Alibaba

Qwen3.7 Plus (International)

Thinking + coding

Input

$0.40/M

Cached

$0.08/M

Output

$1.60/M

$0.000800 per call

standard rate

500 request1.00M context

Data-use warning: Official pricing table confirms a limited-time 20% off label, but no expiry or complete eligibility is established from public documents. No promotional discount applied.

API model=qwen3.7-plus, currently equivalent to qwen3.7-plus-2026-05-26; Singapore / International. Listed standard prices exclude limited-time 20% off labels. Tier selected by TOTAL INPUT tokens per request and applies to all input and output tokens. Same output price in thinking and non-thinking; reasoning tokens billed as output. Implicit cache is in rate; explicit cache creation/read are USD 0.5/0.04 per M at input≤256,000 and 1.5/0.12 per M above 256,000. Maximum input 991,808 / thinking 983,616. Batch File API confirms text support and 50% of realtime price, but batch context is capped at 256K; optional batch rates omitted to keep standard/non-batch scope separate.

Verify provider price ↗

xAI

Grok Build 0.1

Input

$1.00/M

Cached

$0.20/M

Output

$2.00/M

$0.001000 per call

standard rate

500 request256.0K context

Data-use warning: Inclusive long-context boundary follows explicit pricing table and fresh pricing.md despite generic model/cache prose saying 'exceed'. Public HTML contains an internal batchDiscountPercent value for this model, but visible pricing excludes it and model Details says Batch API Not supported; no batch discount is applied.

Exact public API model: grok-build-0.1, an agentic coding/engineering text model, not the Grok Build subscription product. Model page documents aliases grok-code-fast-1, grok-code-fast and grok-code-fast-1-0825; they are not separately priced original models. Batch API is explicitly not supported, so no batch price is included. Not currently served by the US endpoint. Shared cache conditions apply.

Verify provider price ↗

Mistral

Mistral Large 4

Public Preview

Input

$0.70/M

Cached

$0.07/M

Output

$2.10/M

$0.001050 per call

standard rate

Data-use warning: Public Preview permits silent updates. The marketing pricing FAQ's unversioned Mistral Large example is not the Large 4 price. No automatic long-context premium is documented in the inspected rate table.

Direct Mistral global/default Standard Tier text-token prices, USD per 1 million tokens. Card displays context as 1M; exact integer limit not independently documented in the material inspected, so contextTokens is omitted. One documented API identity is retained; latest/snapshot aliases are not separate rows. Cache-hit price is not a cache-write or storage price; see supplier sourceNotes.

Verify provider price ↗

Z.AI

GLM-4.7

Input

$0.60/M

Cached

$0.11/M

Output

$2.20/M

$0.001100 per call

standard rate

500 request200.0K context

Still explicitly documented and in the current direct API enum, not retired. Exact GLM-4.7 identity, not its Flash/FlashX variants. Text-only; docs publish 200K context. Cache new content standard input; hits listed cached-input rate. Storage Limited-time Free, numerical TTL and expiry unspecified.

Verify provider price ↗

Z.AI

GLM-4.6

Input

$0.60/M

Cached

$0.11/M

Output

$2.20/M

$0.001100 per call

standard rate

500 request200.0K context

Still explicitly documented and in the current direct API enum, not retired. Text-only; published 200K context. Cache new content standard input; cache hits listed cached-input rate. Storage Limited-time Free, unspecified numerical TTL and expiry. Legacy does not mean unavailable.

Verify provider price ↗

Alibaba

Qwen3.6 35B A3B (International)

Open-weight small MoE

Input

$0.375/M

Cached

-

Output

$2.25/M

$0.001125 per call

standard rate

500 request262.1K context

API model=qwen3.6-35b-a3b; distinct currently documented small sparse MoE tier, with no newer 35B-A3B family identity listed. Singapore / International. Both thinking/non-thinking text output prices match; reasoning tokens billed as output. Cache unsupported, so cachedInputPerM omitted, not zero. Batch unsupported. Exact context 262,144; maximum input 260,096 / thinking 258,048; output 65,536. Summary pricing uses 0<Token≤256K, where K=1000; model page instead gives exact input capacity. No higher prompt-price tier documented; do not invent one above 256,000.

Verify provider price ↗

Google

Gemini 3.5 Flash-Lite

Budget

Input

$0.30/M

Cached

$0.03/M

Output

$2.50/M

$0.001250 per call

standard rate

500 request1.05M context

Data-use warning: Source-representation discrepancy: the initial ai.google.dev extraction and an earlier GET of aistudio pricing showed Batch cache $0.02, but fresh GETs of aistudio.google.com/docs/pricing.md at 22:20:22 and 22:20:36 GMT showed $0.015. Current raw Markdown value used; not an inferred 50% discount. Cache-hit rate excludes explicit storage and does not price cache writes. Implicit caching is enabled for Gemini 2.5 and newer, but this model's exact minimum is absent from the inspected developer cache table; no minimum guessed. Explicit caching Beta on GenerateContent; default TTL one hour. Tools/grounding are additional.

Exact stable API identity gemini-3.5-flash-lite; baseline calculator ID preserved. Text-scope Standard rate is $0.30 input, $2.50 output including thinking, $0.03 cache-hit per 1M tokens. Explicit-cache storage adds $1.00 per 1M tokens per hour for Standard and Batch. No prompt-length tier or promotion expiry is documented. Batch cached-input rate corrected from baseline $0.02 to $0.015 using repeated fresh GETs of the official migrated pricing Markdown.

Verify provider price ↗

xAI

Grok 4.3

1M context

Input

$1.25/M

Cached

$0.20/M

Output

$2.50/M

$0.001250 per call

standard rate

500 request1.00M context

Data-use warning: Grok 4.3 is not currently available on the US endpoint; do not apply its old 1.1 residency multiplier. Inclusive long-context boundary follows explicit pricing table and fresh pricing.md despite generic model/cache prose saying 'exceed'.

Exact public API model: grok-4.3; grok-4.3-latest is an alias, not another row. Global/default standard text prices and existing batch values verified. Supports configurable reasoning including none. The official batch discount is 20%, not 50%; no expiry is stated. Removed baseline usResidencyMultiplier because the current US endpoint serves only grok-4.7 and grok-4.6. Shared cache conditions apply.

Verify provider price ↗

xAI

Grok 4.20 (Reasoning)

Input

$1.25/M

Cached

$0.20/M

Output

$2.50/M

$0.001250 per call

standard rate

500 request1.00M context

Data-use warning: Inclusive long-context boundary follows explicit pricing table and fresh pricing.md despite generic model/cache prose saying 'exceed'.

Current stable public API alias: grok-4.20-reasoning, explicitly documented as an alias of priced model grok-4.20-0309-reasoning. One row represents this reasoning variant; dated snapshots and latest aliases are not additional rows. Global/default standard text prices; reasoning tokens billed at output rate. Batch supported with official 20% discount. Not currently served by the US endpoint. Shared cache conditions apply.

Verify provider price ↗

xAI

Grok 4.20 (Non-Reasoning)

Input

$1.25/M

Cached

$0.20/M

Output

$2.50/M

$0.001250 per call

standard rate

500 request1.00M context

Data-use warning: Inclusive long-context boundary follows explicit pricing table and fresh pricing.md despite generic model/cache prose saying 'exceed'. The model page also displays a generic reasoning capability label; the exact named non-reasoning identity is retained without relabeling it as the reasoning model.

Current stable public API alias: grok-4.20-non-reasoning, explicitly documented as an alias of priced model grok-4.20-0309-non-reasoning. Distinct non-reasoning model variant; no duplicate dated/latest aliases included. Global/default standard text prices; batch supported with official 20% discount. Not currently served by the US endpoint. Shared cache conditions apply.

Verify provider price ↗

xAI

Grok 4.20 Multi-Agent Beta

Beta

Input

$1.25/M

Cached

$0.20/M

Output

$2.50/M

$0.001250 per call

standard rate

500 request1.00M context

Data-use warning: Beta interface may change; hidden sub-agent token use can materially increase total cost. Tool invocations, if enabled, cost extra and are excluded from these text rates. Inclusive long-context boundary follows explicit pricing table and fresh pricing.md despite generic model/cache prose saying 'exceed'.

Current stable public API alias: grok-4.20-multi-agent, documented as alias of priced model grok-4.20-multi-agent-0309 and used by the official guide. This is a distinct publicly priced beta multi-agent text/reasoning tier, not multiple snapshot rows. Works without built-in tools through Responses API or xAI SDK; Chat Completions is not supported. All leader and sub-agent input, output and reasoning tokens are billed, not only visible final answer tokens. Agent counts change usage, not the published per-token rate. Batch supported per model Details with 20% official discount; unavailable under ZDR. Not currently served by the US endpoint. Shared cache conditions apply.

Verify provider price ↗

Tencent

Hunyuan Hy4 Preview

Preview

Input

$0.834/M

Cached

$0.042/M

Output

$2.501/M

$0.001250 per call

standard rate

Data-use warning: Preview is not GA. Rates are USD per million billed tokens, not the mainland release's CNY prices. Cached-input rate covers actual cache hits only. The pricing table shows '-' for both 5-minute and 1-hour cache-write prices; this is not an explicit zero-price statement. HY-specific cache-write, storage and exact TTL charges/limits are not published in the reviewed documentation, so no free writes/storage, fixed TTL, guaranteed hit or minimum cacheable-prefix length is assumed. Account activation is required; the billing guide states USD 10 is frozen per model when enabling postpaid or claiming trial quota. Trial/subscription benefits are not applied to these rates.

Exact public API parameter is hy4-preview. Tencent identifies it as an early Hunyuan Hy4 iteration, not Hy3, an older Hunyuan Turbo model, or a GA Hy4 release. Scope is Tencent-operated TokenHub pay-as-you-go, standard online non-batch text requests using the Singapore endpoint with Global resource scheduling. Documentation renders context as 1M, maximum input as 960k and maximum output as 64k; numeric contextTokens is omitted because these abbreviated limits are not an exact integer specification. Supports Chat Completions, Responses and Anthropic Messages after account/model activation. No input-length pricing condition or peak/off-peak condition is listed for this model.

Verify provider price ↗

Alibaba

Qwen3.8 27B (International)

Open-weight dense

Input

$0.50/M

Cached

$0.10/M

Output

$3.00/M

$0.001500 per call

standard rate

500 request1.00M context

Exact API model=qwen3.8-27b; current distinct dense-model tier. Singapore / International. Standard text input/output, both thinking and non-thinking. Implicit cache in rate; explicit creation/read USD 0.625/0.05 per M. No documented Global price table, so no Global price inferred from Beijing. Maximum input 991,808 / thinking 983,616; batch unsupported.

Verify provider price ↗

Z.AI

GLM-5

Input

$1.00/M

Cached

$0.20/M

Output

$3.20/M

$0.001600 per call

standard rate

500 request200.0K context

Still explicitly documented and in the current direct API enum, not retired. Stable GLM-5 identity; not substituted for GLM-5-Turbo. Text-only; context published as 200K marketed tokens. Cache new content uses standard input and cache hits use listed rate; storage Limited-time Free, no published numerical TTL or expiry.

Verify provider price ↗

Google

Gemini 3.8 Flash

Promo through 2026-12-31

Input

$0.75/M

Cached

$0.075/M

Output

$3.75/M

$0.001875 per call

standard rate

500 request1.05M context

Data-use warning: Cache-hit rate is not an all-in cache-write/storage price. Implicit caching requires at least 4,096 input tokens and does not guarantee hits. Explicit caching is Beta and available through GenerateContent, not Interactions; default TTL is one hour, with token-size and TTL-dependent storage charges. Developer documentation does not separately specify an explicit cache-creation/write token rate; do not assume it is free or import Cloud-specific billing. Output includes thinking; tools/grounding are extra.

Exact stable API identity: gemini-3.8-flash; baseline calculator ID preserved. Current default paid Standard rates apply through December 31, 2026. Google publishes January 1, 2027 Standard input/output/cache-hit rates of $1.50/$7.50/$0.15 per 1M and Batch rates $0.75/$3.75/$0.075. Explicit-cache storage is an additional $0.50 per 1M cached tokens per hour through December 31, then $1.00 per 1M per hour, in both Standard and Batch tables. No separate prompt-length pricing tier is documented.

Verify provider price ↗

Google

Gemini 3.6 Flash

Previous generation; promo through 2026-12-31

Input

$0.75/M

Cached

$0.075/M

Output

$3.75/M

$0.001875 per call

standard rate

500 request1.05M context

Data-use warning: Cache-hit rate excludes explicit storage and does not establish cache-write pricing. Implicit-cache minimum 4,096 tokens; hits not guaranteed. Explicit caching is Beta on GenerateContent only; default TTL one hour, storage billed by cached-token count and TTL. No distinct explicit-write rate established by the developer docs inspected. Output includes thinking; tool/grounding charges are additional.

Exact stable API identity: gemini-3.6-flash; baseline calculator ID preserved. Google explicitly says this previous-generation model remains available for production use, and no shutdown is announced. Current Standard rates apply through December 31, 2026; January 1, 2027 Standard input/output/cache-hit rates are $1.50/$7.50/$0.15 per 1M, Batch $0.75/$3.75/$0.075. Standard and Batch explicit-cache storage is additional $0.50 per 1M tokens per hour through December 31, then $1.00. No prompt-length price tier is documented.

Verify provider price ↗

DeepSeek

DeepSeek V4 Pro

Peak/off-peak pricing

Input

$1.32/M

Cached

$0.044/M

Output

$3.96/M

$0.001980 per call

standard rate

Data-use warning: The September 10 release article still contains a planned September 14 Pro reroute, but the official changelog explicitly reverses that plan and says V4 Pro service and billing continue after September 14. Current pricing, first-call and chat API reference pages independently continue to list deepseek-v4-pro. Prefer those current documents. Do not mistake the time-based schedules for long-context or fast tiers.

Current stable public API identity deepseek-v4-pro is served by DeepSeek-V4-Pro-0813. Direct standard non-batch text rates in USD per 1M apply to both thinking and non-thinking modes. rate contains the published PEAK schedule as a conservative untimed calculator baseline, not an all-hours flat price. Official OFF-PEAK rates are input/cache-miss $0.66, output $1.98, and cached-input/cache-hit $0.022 per 1M tokens. Peak hours are 01:00-04:00 and 06:00-10:00 UTC Monday-Friday, excluding Chinese public holidays; every other hour, weekends and Chinese public holidays are off-peak. The pricing page states 1M context and 384K maximum output; the chat reference gives maximum output as 393216 and defines context as input plus output. contextTokens is omitted because the checked pages do not specify an exact integer for 1M context. Caching is automatic, disk-based, best-effort and dependent on full persisted-prefix-unit matches; initial/missed input uses the cache-miss rate. Construction takes seconds and idle caches generally expire after hours to days, with no guaranteed TTL. Historical first-party cache documentation says no extra cache-service fees and free storage; the current pricing page lists no separate cache-write/storage premium. No documented long-context token-price uplift, batch schedule, fast surcharge or US-residency multiplier is published on the current price page; those fields are omitted. Pro is currently documented as an active service, not kept merely as a stale legacy row.

Verify provider price ↗

Kimi

Kimi K2.6

Existing users only

Input

$0.95/M

Cached

$0.16/M

Output

$4.00/M

$0.002000 per call

standard rate

500 request262.1K context

Data-use warning: Supplier plans to retire all K2 series models by the end of December 2026. Preserve only for eligible existing-user estimates; not a new-user model recommendation. Batch cache hits are best-effort. Do not substitute the mathematically derived 0.096 for the published batch cache rate 0.10.

Baseline retained as documented legacy, not available to newly registered users from October 31, 2026; existing users remain eligible. Stable API identity is kimi-k2.6; baseline calculator id is preserved. General-purpose text/coding/reasoning model; thinking defaults on but may be disabled, with no separately priced mode row. USD global/default non-batch rates before taxes. K2.6 does not support separately billed cache writes. Official batch table explicitly lists cached input USD 0.10 per 1M tokens; no multiplication of a percentage was used.

Verify provider price ↗

Kimi

Kimi K2.7 Code

Code / existing users only

Input

$0.95/M

Cached

$0.19/M

Output

$4.00/M

$0.002000 per call

standard rate

500 request262.1K context

Data-use warning: Unavailable to newly registered users from October 31, 2026. All K2 series planned for retirement by end December 2026. Batch cache hits are best-effort. Caching guide FAQ abbreviates K2.7 identities as kimi-k2.7/kimi-k2.7-highspeed; model list and pricing table use the full -code identities, which are the identities validated here.

Baseline retained as documented existing-user legacy. Stable API identity: kimi-k2.7-code; baseline calculator id preserved. Direct global/default text/coding API rates, USD per 1M tokens before taxes. Thinking and Preserved Thinking are always on; reasoning contributes to output and historical reasoning sent again contributes to input. No separately billed cache writes. Batch rates are explicit published table entries.

Verify provider price ↗

Meta Muse

Muse Spark 1.3 Standard

Standard

Input

$1.25/M

Cached

$0.15/M

Output

$4.25/M

$0.002125 per call

standard rate

500 request1.05M context

Exact API identity: muse-spark-1.3; retained calculator ID is not the API ID. Latest Spark version, recommended for new work. Standard-tier prompts and completions are not used to train Meta models; Standard is not ZDR. Direct default non-batch text USD rates per 1M tokens. No long-context premium. Automatic prefix cache-hit pricing only; uncached tokens use input pricing. Separate cache-write/storage prices are not published in reviewed docs. Default in-memory cache has no fixed published TTL; optional Responses 24h retention is a hint. Supports max reasoning unlike its Contributor variant.

Verify provider price ↗

Meta Muse

Muse Spark 1.2 Standard

Standard

Input

$1.25/M

Cached

$0.15/M

Output

$4.25/M

$0.002125 per call

standard rate

500 request1.05M context

Exact API identity: muse-spark-1.2; retained calculator ID is not the API ID. Previous Spark version, still explicitly documented and publicly priced on Meta Model API; Legacy denotes previous generation, not retired. Standard-tier prompts and completions are not used to train Meta models; not ZDR. Direct default non-batch text USD per 1M tokens; no long-context premium. Cache-hit rate only; cache misses use uncached input. Cache-write/storage price not separately published. Default in-memory eviction has no fixed documented TTL; optional Responses 24h retention is a hint.

Verify provider price ↗

Meta Muse

Muse Spark 1.1 Standard

Standard

Input

$1.25/M

Cached

$0.15/M

Output

$4.25/M

$0.002125 per call

standard rate

500 request1.05M context

Exact API identity: muse-spark-1.1; retained calculator ID is not the API ID. Original Spark version remains publicly documented and priced on Meta Model API, Standard only; Legacy does not mean retired. Not available in Muse Code. Standard-tier prompts and completions are not used to train Meta models; not ZDR. Direct default non-batch text USD per 1M tokens, no long-context premium. Automatic cache-hit rate only; uncached input is priced normally. No separately published cache-write/storage rate or fixed default TTL in reviewed docs; optional Responses 24h retention is only a hint.

Verify provider price ↗

Z.AI

GLM-5.3

Latest flagship

Input

$1.40/M

Cached

$0.26/M

Output

$4.40/M

$0.002200 per call

standard rate

500 request1.00M context

Standard direct API, non-batch text scope. Text-only input and text output; reasoning is always enabled, with low/high/max effort. Context is published as 1M, a marketed capacity rather than a separately documented exact integer boundary. Implicit cache: new content is charged at standard input rates; the listed cache rate applies to cache-hit tokens. Cached Input Storage is 'Limited-time Free'; no expiry date, numerical TTL, guaranteed cache lifetime, minimum cache length, or separate cache-write surcharge is published in the checked pricing/cache docs. This does not establish permanently free cache writes/storage. No automatic long-context surcharge, batch discount, or residency multiplier is documented.

Verify provider price ↗

Z.AI

GLM-5.2

Input

$1.40/M

Cached

$0.26/M

Output

$4.40/M

$0.002200 per call

standard rate

500 request1.00M context

Still explicitly documented and in the current direct API enum, not retired. Legacy denotes a predecessor to GLM-5.3, not unsupported status. Stable model code retained, not a dated snapshot alias. Text-only; published context 1M. Implicit-cache new content uses standard input rate; cache hits use listed rate. Storage Limited-time Free, no published numerical TTL/expiry. No documented long-context price step.

Verify provider price ↗

Z.AI

GLM-5.1

Input

$1.40/M

Cached

$0.26/M

Output

$4.40/M

$0.002200 per call

standard rate

500 request200.0K context

Still explicitly documented and in the current direct API enum, not retired. Legacy is a relative-generation label. Stable API identity, not a snapshot alias. Text-only; docs advertise 200K context, represented here as 200000 marketed tokens, not a claim of an exact 202752/204800 boundary. Cache new content is standard input; hits use listed rate. Storage Limited-time Free, unspecified expiry and numerical TTL.

Verify provider price ↗

Mistral

Z.ai GLM 5.3 (Mistral-hosted)

GA · Third-party

Input

$1.40/M

Cached

$0.14/M

Output

$4.40/M

$0.002200 per call

standard rate

Z.ai model hosted by Mistral without modifications; provider identifies the direct API supplier, not model authorship. USD per 1 million text tokens on Mistral's global/default Standard Tier, not Z.ai's own API pricing. Card displays 1M context and 128k max output; exact integer limits not established. See sourceNotes for cache conditions.

Verify provider price ↗

OpenAI

GPT-5.4 Mini

Input

$0.75/M

Cached

$0.075/M

Output

$4.50/M

$0.002250 per call

standard rate

500 request400.0K context

Active separately priced mini coding/subagent model; maximum input 272,000/output 128,000. No long tier published for this model. Earlier-model implicit caching with no additional write charge; do not assume GPT-5.6 30m/12h write billing applies.

Verify provider price ↗

Alibaba

Qwen3.8 Max (Global)

Global scope

Input

$1.65/M

Cached

$0.206/M

Output

$4.951/M

$0.002475 per call

standard rate

500 request1.00M context

Calculator id is scope-qualified; API model remains qwen3.8-max, not qwen3.8-max-global. Global scope at Frankfurt, Virginia, Tokyo or Hong Kong; inference may execute in Chinese mainland or overseas. Same standard thinking/non-thinking text output price. No prompt-price tiers. cachedInputPerM is implicit cache; explicit read USD 0.137/M, creation USD 2.063/M. Maximum input 991,808 non-thinking / 983,616 thinking. Global batch unsupported. US-only residency is a different scope priced at USD 2 input / 6 output / 0.25 implicit cache per M; no single exact residency multiplier is assumed.

Verify provider price ↗

OpenAI

GPT-6.1 Terra

Latest Terra

Input

$1.00/M

Cached

$0.05/M

Output

$5.00/M

$0.002500 per call

standard rate

500 request1.05M context

Data-use warning: The supplied 30% reseller discount is unverified and is not applied. Cache writes must be priced separately from cache hits.

Requested exact public API identity confirmed. Text generation and reasoning; Responses and Chat Completions supported; tool calling requires Responses. Maximum input 922,000/output 128,000. Cache writes USD 1.25/1M for 30m or 2.50/1M for 12h; long-context writes 2.50/5. The >272,000-input-token surcharge applies to the full request. Flex shares Batch rates; Fast long-context rates 4 input/15 output/0.20 cached per 1M. US and EU residency supported, including Fast EU processing.

Verify provider price ↗

Anthropic

Claude Haiku 4.5

Legacy

Input

$1.00/M

Cached

$0.10/M

Output

$5.00/M

$0.002500 per call

standard rate

500 request200.0K context

Data-use warning: Corrected baseline contextTokens from 1000000 to 200000 and removed unsupported usResidencyMultiplier=1.1. Sending inference_geo to Haiku 4.5 returns a 400 error. Cache-hit cost alone does not price cache creation.

Preserved baseline legacy model with corrected stable API snapshot: baseline claude-haiku-4-5 remains a documented convenience alias resolving to claude-haiku-4-5-20251001; only the pinned identity is listed. Official context is 200,000 tokens, not baseline 1M. Direct API status is active (legacy), still available despite the earliest retirement commitment date having passed. Standard global/default text rates in USD per million tokens; standard and batch rates revalidated. Cache writes $1.25/MTok for 5m and $2/MTok for 1h, minimum cacheable prefix 4,096 tokens. Batch cache hits $0.05/MTok follow the documented 50% all-usage discount. No first-party US-residency multiplier: this model predates Claude 4.6 and does not support inference_geo.

Verify provider price ↗

xAI

Grok 4.7

Frontier

Input

$2.00/M

Cached

$0.50/M

Output

$6.00/M

$0.003000 per call

standard rate

500 request500.0K context

Data-use warning: The authoritative pricing table and its fresh Markdown explicitly use >=200,000 prompt tokens, encoded here as inclusive. Model-page prose and cache usage prose instead say 'exceed'; this documentation inconsistency remains flagged. Grok 4.7 Fast is not a public API model and is excluded.

Exact public API model: grok-4.7. Global/default standard non-batch text rates. Latest flagship for coding and general work; configurable reasoning. All tokens in a request use long-context rates once total prompt tokens, including cached tokens, reach the threshold. Batch is supported with supplier-confirmed 20% discount. US endpoint is supported; the 1.1 multiplier is separate from the global base rates. Cache rates cover hits only; see shared cache and retention sourceNotes.

Verify provider price ↗

xAI

Grok 4.6

Frontier

Input

$2.00/M

Cached

$0.50/M

Output

$6.00/M

$0.003000 per call

standard rate

500 request500.0K context

Data-use warning: Inclusive long-context boundary follows the explicit pricing table and fresh pricing.md, despite generic model/cache prose saying 'exceed'.

Exact public API model: grok-4.6. Still documented and priced, so the baseline row is retained. Global/default text rates; batch supported with official 20% discount. US endpoint supported with separate 1.1 token premium. Reasoning is charged at output rate. Shared cache and region conditions are in sourceNotes.

Verify provider price ↗

xAI

Grok 4.5

Input

$2.00/M

Cached

$0.30/M

Output

$6.00/M

$0.003000 per call

standard rate

500 request500.0K context

Data-use warning: Inclusive long-context boundary follows the explicit pricing table and fresh pricing.md, despite generic model/cache prose saying 'exceed'.

Exact public API model: grok-4.5. Coding/engineering and reasoning model with a genuinely distinct cached-input rate from Grok 4.6/4.7. Aliases grok-4.5-latest and grok-build-latest are not separate priced rows. Batch supported with official 20% discount. Not currently served by the US endpoint. Global/default standard text scope; shared cache conditions apply.

Verify provider price ↗

Alibaba

Qwen3.8 Max (International)

Thinking + coding

Input

$2.00/M

Cached

$0.25/M

Output

$6.00/M

$0.003000 per call

standard rate

500 request1.00M context

API model=qwen3.8-max; Singapore endpoint, International inference scope excluding Chinese mainland. Standard text-only rates; same output price for thinking and non-thinking, with chain-of-thought tokens billed as output. No prompt-price tiers. cachedInputPerM is implicit cache; explicit cache read is USD 0.17/M and creation USD 2.5/M. Context is 1,000,000; maximum input 991,808 non-thinking / 983,616 thinking. Text-only batch is not supported in this scope.

Verify provider price ↗

Alibaba

Qwen3.8 2.4T A95B (International)

Open-weight flagship

Input

$2.00/M

Cached

$0.25/M

Output

$6.00/M

$0.003000 per call

standard rate

500 request1.00M context

Exact API model=qwen3.8-2.4t-a95b, a distinct open-source flagship API identity; not relabeled as qwen3.8-max. Singapore / International only for the included scope; no documented Global USD table, so no Global row inferred from Beijing. Standard text input/output, both thinking and non-thinking. Implicit cache USD 0.25/M; explicit creation/read USD 2.5/0.17 per M. Maximum input 991,808 / thinking 983,616. Batch unsupported.

Verify provider price ↗

Kimi

Kimi K2.7 Code HighSpeed

Fast code / existing users only

Input

$1.90/M

Cached

$0.38/M

Output

$8.00/M

$0.004000 per call

standard rate

500 request262.1K context

Data-use warning: Unavailable to newly registered users from October 31, 2026; K2 series planned for retirement by end December 2026. Do not infer batch pricing from the standard K2.7 Code tier.

Baseline retained as documented existing-user legacy and genuinely distinct priced speed tier. Stable API identity: kimi-k2.7-code-highspeed; baseline calculator id preserved. USD global/default direct text/coding API rates before taxes. Same model intelligence and thinking behavior as K2.7 Code; supplier reports approximately 180 output tokens/s, up to 260 in short contexts. No separately billed cache writes. No batch rate listed for this identity.

Verify provider price ↗

OpenAI

GPT-6.1 Sol

Latest Sol

Input

$2.00/M

Cached

$0.10/M

Output

$10.00/M

$0.005000 per call

standard rate

500 request1.05M context

Data-use warning: The supplied 15% reseller discount is unverified and is not applied. Cache-hit pricing alone omits cache-write cost.

Requested exact identity confirmed, distinct from gpt-6-sol. Public Responses and Chat Completions; tool calling requires Responses. Maximum input 922,000 and output 128,000. Cache writes USD 2.50/1M for 30m or 5/1M for 12h; long-context writes 5/10. Long-context rates cover the full API request above 272,000 input tokens. Flex shares published Batch pricing. Fast long-context rates are 8 input/30 output/0.40 cached per 1M. Ultrafast is separately priced at 6x Standard, not a discounted Standard rate. US and EU residency supported; Fast supports EU. Subscription usage in Codex/ChatGPT Work is not this API price scope.

Verify provider price ↗

OpenAI

GPT-6 Sol

Previous Sol

Input

$2.00/M

Cached

$0.20/M

Output

$10.00/M

$0.005000 per call

standard rate

500 request1.05M context

Baseline retained as its own currently documented API identity; do not relabel as GPT-6.1 Sol. Model page directs users to newer 6.1 Sol. Cache writes USD 2.50/1M for 30m or 5/1M for 12h; long writes 5/10. Full-request long tier above 272,000 input tokens. Maximum input 922,000/output 128,000. Flex shares Batch rates. EU residency supports Standard/Flex/Batch but not Fast. Fast long rates input/output/cached 8/30/0.80 per 1M.

Verify provider price ↗

OpenAI

GPT-4o

Deprecated

Input

$2.50/M

Cached

$1.25/M

Output

$10.00/M

$0.005000 per call

standard rate

500 request128.0K context

Data-use warning: Deprecated, with API shutdown January 14, 2027 for currently served August/November 2024 snapshots. The older May-2024 snapshot shut down October 23, 2026 and is not represented here. Recheck/remove before scheduled shutdown.

Published baseline legacy retained because current official pricing and API model page still document it. Stable alias maps to gpt-4o-2024-08-06; no separate dated alias row. Responses/Chat Completions/Batch supported today. Maximum output 16,384. Earlier implicit cache has no extra write charge; in_memory entries generally 5–10 minutes inactive, up to one hour. No published Batch cached rate ('-'), so omitted rather than assumed zero.

Verify provider price ↗

Anthropic

Claude Sonnet 5.5

Balanced

Input

$2.00/M

Cached

$0.10/M

Output

$10.00/M

$0.005000 per call

standard rate

500 request1.00M context

Added the latest current Sonnet tier. Public Claude API model; standard global/default synchronous text rates in USD per million tokens. Cache read price was reduced to $0.10/MTok on October 7, 2026; do not use the earlier $0.20/MTok launch rate. Cache hits do not price cache writes: standard 5m writes $2.50/MTok and 1h writes $4/MTok; minimum cacheable prefix 512 tokens. Full 1M context is standard-priced. Optional US-only inference multiplies all token categories by 1.1. Batch cache hit is $0.05/MTok under the documented 50% all-usage discount.

Verify provider price ↗

Anthropic

Claude Sonnet 5

Legacy

Input

$2.00/M

Cached

$0.20/M

Output

$10.00/M

$0.005000 per call

standard rate

500 request1.00M context

Preserved documented baseline legacy model: direct API active (legacy), with migration to Sonnet 5.5 recommended. Standard global/default text rates in USD per million tokens. $2/$10 input/output prices are now permanent standard prices: the announced introductory period ended August 31, 2026, but the scheduled September 1 price increase was canceled. Cache writes remain $2.50/MTok for 5m and $4/MTok for 1h; cache-hit rate $0.20/MTok is distinct from Sonnet 5.5's $0.10. Minimum cacheable prefix 1,024 tokens. Full 1M context has no long-context premium. US-only inference is optional 1.1x across all token categories. Official batch cache hit is $0.10/MTok.

Verify provider price ↗

Google

Gemini 4 Argon

Intro price through 2026-11-30

Input

$2.00/M

Cached

$0.10/M

Output

$10.00/M

$0.005000 per call

standard rate

500 request1.05M context

Data-use warning: Implicit caching only; explicit cache writes/storage are unsupported. Minimum implicit-cache input is 4,096 tokens; hits are not guaranteed. In multi-turn/agent loops, the full previous request is cached only when the new request adds at least 2,050 tokens; otherwise its last 2,050 tokens are reprocessed at the standard input rate. Output pricing includes thinking tokens. All processed tokens, including internal grounding tool instructions and function declarations, are billed. Grounding/tool charges are additional. No Free-tier access.

Exact stable API identity: gemini-4-argon. Generally available since October 13, 2026; paid Gemini Developer API, not an invite-only or managed-agent price. Standard introductory USD/1M input/output/cache-hit rates apply through November 30, 2026; published December 1 rates are $4/$20/$0.20. Published Batch rates thereafter are $2/$10/$0.10. The 1,048,576-token context window is unified: input, thinking, and output share it. Synchronous output defaults to 65,536 tokens; long decode can use the remaining unified window. No prompt-length price tier is documented.

Verify provider price ↗

Kimi

Kimi K3.1 Pro

1M context

Input

$2.00/M

Cached

$0.20/M

Output

$10.00/M

$0.005000 per call

standard rate

500 request1.05M context

Data-use warning: The supplied USD 1.40 input / 7.00 output / 0.14 cache-hit offer is unverified and is not used. Cache-hit pricing alone does not price cache writes; 1h writes need separate calculation. Batch caching is best-effort, not guaranteed.

Stable API identity: kimi-k3.1-pro. Direct global/default text API, pay-as-you-go, before taxes. Available after a successful minimum USD 1 top-up; not a membership or Kimi Code plan. Flat prices throughout the context window, with no long-context surcharge. Thinking is always enabled; output usage includes reasoning. Implicit prefix caching defaults to 5m TTL: cache writes cost USD 2.00 per 1M tokens for 5m or USD 4.00 for 1h. Batch writes cost USD 1.20 for 5m or USD 2.40 for 1h per 1M tokens. Cache reads, writes and uncached remainder are mutually exclusive input categories, not additive charges on the same tokens; hits renew the original TTL without another write fee.

Verify provider price ↗

Cohere

Command A

Input

$2.50/M

Cached

-

Output

$10.00/M

$0.005000 per call

standard rate

500 request256.0K context

Data-use warning: No official cached-input rate or cache-write, TTL, storage or eligibility schedule was located for this direct API model; cachedInputPerM is omitted, not zero. Tier is a calculator classification, not a supplier billing tier.

Direct Cohere production Chat API, standard text input/output, published dollar prices per 1M tokens. Maximum output 8,000 tokens. Distinct from Command A+ and Command A Reasoning. No regional, batch or long-context price adjustment was documented in the checked standard pricing sources.

Verify provider price ↗

Cohere

Command R+ (08-2024)

Live older generation

Input

$2.50/M

Cached

-

Output

$10.00/M

$0.005000 per call

standard rate

500 request128.0K context

Data-use warning: Pricing-page FAQ groups this version under existing-customer legacy pricing, while its dedicated model page has a public pricing card and the API catalog marks it Live. Availability is based on the dedicated page and catalog; no paid call was made. No official cache rates or cache-write/TTL/storage schedule were located.

Distinct live August 2024 API model, not the deprecated command-r-plus alias for April 2024. Direct standard text Chat API; maximum output 4,000. The Legacy calculator tier reflects an older generation and the supplier recommendation to use Command A, not API retirement.

Verify provider price ↗

OpenAI

GPT-5.6 Terra

Input

$2.00/M

Cached

$0.20/M

Output

$12.00/M

$0.006000 per call

standard rate

500 request1.05M context

Active, separately priced general text/reasoning tier, not an alias of 6.1 Terra. Maximum input 922,000/output 128,000. Cache writes USD 2.50/1M for 30m or 5/1M for 12h; long writes 5/10. Long-context rates cover full request above 272,000 input tokens. Flex shares Batch pricing. Fast long input/output/cached 8/36/0.80 per 1M.

Verify provider price ↗

Google

Gemini 3.1 Pro Preview

Preview; long-context tier

Input

$2.00/M

Cached

$0.20/M

Output

$12.00/M

$0.006000 per call

standard rate

500 request1.05M context

Data-use warning: Preview, not GA; no shutdown date announced. No Free-tier API access. Output prices include thinking tokens. Implicit-cache minimum 4,096 tokens, hits not guaranteed. Explicit caching is Beta through GenerateContent only; default TTL one hour. Cache-hit numbers exclude storage and do not establish an explicit cache-write rate. Cached tokens count in prompt/context limits; non-cached inputs and tools are separately charged.

Exact publicly documented preview API identity gemini-3.1-pro-preview; no invented stable gemini-3.1-pro alias. Baseline calculator ID preserved. Base prices apply when prompt size is <=200,000 tokens; strictly >200,000 prompt tokens selects the higher input, output, and cache-hit prices for that request. Batch cache-hit rates are explicitly the same as Standard, not automatically discounted. Explicit-cache storage adds $4.50 per 1M cached tokens per hour in both context tiers and both Standard and Batch. The customtools endpoint is documented at the same prices and is not duplicated as a distinct priced tier.

Verify provider price ↗

OpenAI

GPT-5.2

Input

$1.75/M

Cached

$0.175/M

Output

$14.00/M

$0.007000 per call

standard rate

500 request400.0K context

Active previous flagship, not tagged deprecated in current full catalog. Maximum output 128,000. No long-context surcharge published for this 400,000-context identity. Earlier implicit caching has no additional write charge; model is listed as supporting in_memory and extended 24h retention, subject to organization settings. Do not apply post-March-5-2026 model-release residency uplift automatically to this December-2025 family.

Verify provider price ↗

OpenAI

GPT-5.4

Input

$2.50/M

Cached

$0.25/M

Output

$15.00/M

$0.007500 per call

standard rate

500 request1.05M context

Active older general coding/reasoning family; maximum output 128,000. >272,000-input long tier applies to full session for Standard/Batch/Flex. No additional cache-write charge; supports older in_memory/24h retention. Batch short cached rate is published as USD 0.13/1M (table rounding preserved, not replaced by inferred 0.125). Fast long cells are unpriced; no Fast long rate inferred.

Verify provider price ↗

Kimi

Kimi K3

1M context

Input

$3.00/M

Cached

$0.30/M

Output

$15.00/M

$0.007500 per call

standard rate

500 request1.05M context

Data-use warning: Cache-hit rates alone do not price TTL writes; do not also charge regular input on the same cache-write tokens. Batch cache hits are best-effort.

Baseline retained: all three published token rates and context remain documented. Stable API identity: kimi-k3, previous-generation flagship still available through direct global/default text API after minimum USD 1 top-up. No context-length pricing tiers. Thinking is always on and contributes to output usage. Cache writes are USD 3.00 per 1M tokens for default 5m TTL or USD 6.00 for 1h; batch writes USD 1.80/3.60 respectively. Cache-hit tokens do not also incur a write charge. Cache reads, writes and uncached remainder are mutually exclusive input categories.

Verify provider price ↗

OpenAI

GPT-5.6 Sol

Promo

Input

$4.00/M

Cached

$0.40/M

Output

$20.00/M

$0.0100 per call

standard rate

500 request1.05M context

Data-use warning: Official promotional rates are already the listed Standard rates, available at least through November 21, 2026. No additional discount applied and no post-November-21 rate guessed; recheck then.

Baseline model retained; exact API identity uses gpt-5.6-sol, not baseline UI slug gpt-5-6-sol. gpt-5.6 is a routing alias, not a separately priced row. Public general text/reasoning model with Responses and Chat Completions supported. Cache writes USD 5/1M for 30m or 10/1M for 12h; long writes 10/20. Full-request long rates above 272,000 input tokens. Flex shares Batch rates. Fast long input/output/cached 16/60/1.60 per 1M. Maximum input 922,000/output 128,000.

Verify provider price ↗

Anthropic

Claude Opus 5.5

Frontier

Input

$4.00/M

Cached

$0.20/M

Output

$20.00/M

$0.0100 per call

standard rate

500 request1.00M context

Data-use warning: Corrected the baseline note: Opus 5.5 fast mode is stable, not research preview. Fast mode cannot be combined with Batch API or Priority Tier commitments. Supplied reseller discount is not verified. Cache write/TTL costs remain separate.

Baseline identity and all token rates revalidated. Public Claude API model with always-on adaptive thinking. Standard global/default synchronous text rates in USD per million tokens. Cache hits are not cache writes: standard 5-minute writes $5/MTok; 1-hour writes $8/MTok; minimum cacheable prefix 512 tokens. Full 1M context has no long-context premium. Fast mode uses the same model with speed=fast, is now stable on Opus 5.5, and is not a separate model row. Fast cache hit $0.40/MTok follows the documented 0.05x multiplier on $8/MTok fast input; fast cache writes are $10/MTok for 5m and $16/MTok for 1h. US-only inference multiplies all token categories by 1.1. Batch cache hit rate is $0.10/MTok under the documented 50% all-usage discount.

Verify provider price ↗

Kimi

Kimi K3.1 Pro Highspeed

Fast / 1M context

Input

$4.00/M

Cached

$0.20/M

Output

$20.00/M

$0.0100 per call

standard rate

500 request1.05M context

Data-use warning: Do not double an entire standard-model bill to estimate Highspeed: the cache-hit rate remains unchanged, and cache writes require TTL-specific accounting. No unsupported batch rate or residency multiplier has been inferred.

Stable API identity: kimi-k3.1-pro-highspeed. Separately priced speed tier, same intelligence as K3.1 Pro; official model guide reports about 2–3 times the speed. USD global/default direct text API rates before taxes; minimum USD 1 top-up required. Flat context pricing. Implicit cache-write rates: USD 4.00 per 1M tokens for default 5m TTL and USD 8.00 for 1h. Cached input is USD 0.20, not twice the standard variant's cache-hit rate. Cache reads, writes and uncached tokens are mutually exclusive categories. No batch rate is listed for this identity.

Verify provider price ↗

Meta Muse

Muse Wonder 1.0 Standard

Preview · Standard

Input

$4.00/M

Cached

$0.20/M

Output

$20.00/M

$0.0100 per call

standard rate

500 request1.05M context

Exact API identity: muse-wonder-1.0. Public Meta Model API pay-as-you-go preview, Standard mode only. Direct default non-batch text rates in USD per 1M tokens. Standard content is not used to train Meta models; this is not zero data retention. No long-context premium. Cache-hit rate applies only to reported cached input tokens; automatic cache misses/first prefixes consume uncached input. Separate cache-write or retention/storage rates are not published in the reviewed pricing/cache docs; do not treat the cache-hit rate as a cache-write price. Default in-memory retention has no fixed documented TTL; optional Responses 24h retention is an evictable hint. Model docs say unavailable in South Korea. Not Muse Wonder Full.

Verify provider price ↗

Anthropic

Claude Opus 5

Legacy

Input

$5.00/M

Cached

$0.50/M

Output

$25.00/M

$0.0125 per call

standard rate

500 request1.00M context

Data-use warning: Retained only because it is a documented baseline legacy row. Fast mode has access conditions and cannot combine with Batch API or Priority Tier. Cache writes are not covered by the cache-hit rate.

Preserved documented baseline legacy model: direct API active (legacy), with migration to Opus 5.5 recommended. Standard global/default text rates in USD per million tokens. 5m cache writes $6.25/MTok, 1h writes $10/MTok, minimum prefix 512 tokens. Full 1M context at standard pricing. Optional US-only inference 1.1x applies across categories. Fast mode remains research preview, requires fast-mode-2026-02-01 beta header and account-manager access or waitlist; not a broadly guaranteed default mode. Fast cache hit $1/MTok follows documented 0.1x of $10/MTok fast input; fast cache writes $12.50/MTok for 5m and $20/MTok for 1h. Batch cache hit is $0.25/MTok under the official 50% all-usage discount.

Verify provider price ↗

OpenAI

GPT-5.5

Input

$5.00/M

Cached

$0.50/M

Output

$30.00/M

$0.0150 per call

standard rate

500 request1.05M context

Active older distinct text/reasoning family; stable API alias rather than additional dated snapshot row. Maximum output 128,000. Long tier applies for the full session above 272,000 input tokens for Standard/Batch/Flex. No additional cache-write charge; only implicit caching; prompt_cache_retention supports 24h only (typically around 30m, up to 24h). Fast short-context table is shown; its long-context cells are unpriced, so no Fast long rate inferred.

Verify provider price ↗

OpenAI

Chat Latest

Input

$5.00/M

Cached

$0.50/M

Output

$30.00/M

$0.0150 per call

standard rate

500 request400.0K context

Data-use warning: OpenAI does not recommend this moving ChatGPT model for production API use and instead recommends GPT-6 Astra. Numeric cached-input price alone does not establish cache-write cost or retention eligibility.

Distinct priced public API identity for the latest ChatGPT Instant model. Responses and Chat Completions supported; Batch unsupported. Maximum input 272,000/output 128,000. No long-context tier published. Underlying snapshot changes regularly. Cache-write/TTL rules cannot safely be inferred from this moving alias's name; confirm applicable model behavior before estimating writes.

Verify provider price ↗

OpenAI

GPT-6 Astra

Flagship

Input

$10.00/M

Cached

$1.00/M

Output

$50.00/M

$0.0250 per call

standard rate

500 request1.05M context

Data-use warning: cachedInputPerM prices cache hits only, not cache writes. Residency/FedRAMP premiums are not included in global rates; see sourceNotes.

Baseline retained. Public Responses and Chat Completions text API. Maximum input 922,000; maximum output 128,000. Standard global rates apply through 272,000 input tokens; long-context rates apply to the full request above that threshold. Cache writes, not reads: USD 12.50/1M for default 30m minimum lifetime or 25/1M for 12h; long-context writes 25/50. Flex has the published Batch rates, with slower responses and occasional resource unavailability. Fast prices shown are short-context only; its long-context table is 40 input/150 output/4 cached per 1M. Fast has no EU data residency and no latency SLA for Astra. Ultrafast is a separate published mode, not Standard.

Verify provider price ↗

Anthropic

Claude Fable 5.1

Frontier

Input

$10.00/M

Cached

$0.25/M

Output

$50.00/M

$0.0250 per call

standard rate

500 request1.00M context

Data-use warning: General public access is distinct from Claude Mythos 5.1 verification-only access. Safeguards can restrict some cybersecurity and biology workloads. Cache write/TTL costs must be accounted for separately.

Baseline identity, standard rates, batch rates and context revalidated. Public Claude API model; adaptive thinking is always on. Standard global/default synchronous text rates in USD per million tokens. cachedInputPerM prices cache hits/refreshes, not writes: 5-minute writes cost $12.50/MTok and 1-hour writes $20/MTok. Minimum cacheable prefix is 512 tokens. Full 1M context has no long-context premium. US-only inference is an optional 1.1x modifier on every token category, not the default. Batch cache rate follows the documented 50% discount on all usage; cache hits are best effort.

Verify provider price ↗

OpenAI

GPT-5.2 Pro

Input

$21.00/M

Cached

-

Output

$168.00/M

$0.0840 per call

standard rate

500 request400.0K context

Data-use warning: Official pricing Markdown lists Batch rates, but current model endpoint support marks Batch Not supported. Batch fields omitted because applicability conflicts.

Active previous Pro tier; stable API alias, no duplicate dated snapshot. Responses supported, Chat Completions unsupported; maximum output 128,000. Cached input is '-' in pricing and omitted, not zero. No published long tier for this context size.

Verify provider price ↗

OpenAI

GPT-5.5 Pro

Pro

Input

$30.00/M

Cached

-

Output

$180.00/M

$0.0900 per call

standard rate

500 request1.05M context

Active, distinct higher-compute Pro tier. Responses and Batch supported; Chat Completions unsupported. Maximum output 128,000. No cached-input discount; cache rate omitted, not zero. No additional cache-write charge; 24h retention only. Explicit pricing-document context footnote supplies the >272,000-input threshold for Standard long rates. Batch long-context cells are '-' and not inferred.

Verify provider price ↗

OpenAI

GPT-5.4 Pro

Input

$30.00/M

Cached

-

Output

$180.00/M

$0.0900 per call

standard rate

500 request1.05M context

Data-use warning: Official pricing Markdown lists Batch rates, but the current model page marks Batch Not supported. Batch fields omitted because applicability conflicts; no inference call made to resolve it.

Active distinct Pro tier; Responses supported, Chat Completions unsupported. Maximum output 128,000. No cached-input rate published; omit rather than substitute zero. >272,000-input-token long tier applies to full session. Earlier models have no additional cache-write charge.

Verify provider price ↗

Selected model comparison

This table prices the tokens currently in the calculator. With the default empty prompt it shows $0 input. It is not the November 2, 2026 list price.

ModelInputOutputTotal / callvs lowest

xAI

Grok 4.3

Training-data warning

$0.00$0.001250$0.001250Lowest

Google

Gemini 3.8 Flash

Training-data warning

$0.00$0.001875$0.0018751.5×

Meta Muse

Muse Spark 1.3 Standard

$0.00$0.002125$0.0021251.7×

Anthropic

Claude Sonnet 5.5

$0.00$0.005000$0.0050004.0×

Kimi

Kimi K3.1 Pro

Training-data warning

$0.00$0.005000$0.0050004.0×

OpenAI

GPT-5.2

$0.00$0.007000$0.0070005.6×

Share comparison

Create a team-ready URL for this model shortlist and its pricing settings. To protect sensitive work, shared links never include the prompt text you pasted above.

The message includes a link back to the free AI Answers score for teams that want help turning model costs into an accountable AI plan.

How to use this estimate responsibly

  1. Paste a representative prompt and set the response length you expect.
  2. Compare the quality, latency and context needs of a small model shortlist; do not choose on price alone.
  3. Turn on cache, batch, fast mode, US inference or off-peak pricing only if your actual provider configuration uses that documented feature or schedule.
  4. Validate tokenization, taxes, storage, grounding, modality and contract charges separately before approval.
Credits, not token API

Manus agent credits

Manus cost is credit-based for agent tasks rather than a stable public dollars-per-million-token API price. Use the invitation to trial research, content and analysis workflows without inventing a token conversion.

API text tokensManus credits
What the page says it metersPrompt and response text, priced per million tokensAgent tasks, which can combine research, browser work, code, files and model use
Dollar rate on this pageThe table above, dated November 2, 2026None. Do not invent one.
How to try itProvider pricing pages linked in the existing FAQThe Manus invitation already on this page. No credit card under the current invitation flow, as the FAQ states.

A company-level question about revenue, compute exposure and cash is a different tool from a single API call: AI economics scorecard.

Claim free Manus credits →

https://manus.im/invitation/QPAHJBYCPQ3IK?utm_source=invitation&utm_medium=social&utm_campaign=system_share

Frequently asked questions about AI token costs

These answers match the FAQPage schema on this page and distinguish documented provider rates from planning assumptions.

First-party pricing sources

The model module centralizes the source date and provider page for every displayed API row. Provider schedules, preview availability, promotions and terms can change.

Verified: .

Last amended: .

Claim free Manus credits →

Related services

Need expert help beyond the tool?

Agentic AI Lead Generation

Plan autonomous AI workflows with commercial, data and measurement controls.

AI Marketing Strategy

Connect model, workflow and channel choices to a measurable operating plan.

AEO, GEO & AI Search

Build content that can be understood, evaluated and cited by AI answer engines.

From the blog

Go deeper with expert analysis

AI Pricing

GPT-5.6 Luna 80% Price Cut: Does It Make Agentic Marketing Economically Viable?

A commercial framework for comparing model price changes against workflow quality and governance needs.

AI Models

Is Kimi K3 Really Open Enough to Challenge ChatGPT, Claude and Gemini?

A practical view of model evaluation beyond an individual API price row.

Agentic AI

Agentic AI Lead Generation: The Cost-Per-Resolution Model Explained

Connect individual model estimates to end-to-end workflow economics.

View all articles →

More free tools

📊AI Economics Scorecard✨AI Prompt Improver🔗UTM Campaign BuilderView all tools →