A token is the unit of text an AI model processes rather than a fixed number of characters or words. In ordinary English, a practical rough estimate is about four characters per token, but punctuation, code, non-English text and each provider's tokenizer can materially change that result. Use the provider tokenizer when a deployment decision needs an exact count.
Free AI planning tool
AI Token & Cost Calculator: verified API price estimates
Compare current text-token rates across 31 verified model rows from ChatGPT and GPT API models, Claude, Gemini, Grok, Kimi and Muse. Manus is shown separately as a credit-based agent product, not as a made-up token API rate.
31 verified model rows6 API providersCache, batch and long-context rules where documented
Need a different starting point? Browse all free AI marketing tools.
Standard text rates as of 28 September 2026
These are the list prices already printed on the model cards, in US dollars per million text tokens. They are not a quote, not a per-call total, and not a GBP budget. The comparison lower down prices whatever text is in the calculator. An empty prompt shows $0 input. Verify the provider page before you spend.
| Provider | Model | Input /M | Cached input /M | Output /M | Condition already on this page |
|---|---|---|---|---|---|
| Meta Muse | Muse Spark 1.3 Contributor | $0.10 | $0.002 | $0.20 | Contributor terms allow Meta to use prompts and completions to train future Meta models. |
| Meta Muse | Muse Spark 1.2 Contributor | $0.10 | $0.002 | $0.20 | Same Contributor data-use warning. |
| Gemini 2.5 Flash-Lite | $0.10 | $0.01 | $0.40 | Text, image and video input rate. Audio input is a separate published rate, not in this grid. | |
| OpenAI | GPT-6 Luna | $0.10 | $0.01 | $0.50 | A long-context rate is published, but the cited docs do not state a safe automatic threshold. |
| OpenAI | GPT-4o mini | $0.15 | $0.07 | $0.60 | Legacy. Context on the card: 128.0K. |
| Gemini 3.1 Flash-Lite | $0.25 | $0.03 | $1.50 | Text, image and video input rate. Audio is separate. | |
| Gemini 3.5 Flash-Lite | $0.30 | $0.03 | $2.50 | Budget row. No extra condition on the card. | |
| xAI | Grok 4.3 | $1.25 | $0.20 | $2.50 | Context on the card: 1.00M. Documented batch discount is 20% for this model, not 50%. |
| Gemini 3.8 Flash | $0.75 | $0.07 | $3.75 | Promo through 2026. Published standard price becomes $1.50 input and $7.50 output from 1 January 2027. | |
| Gemini 3.7 Flash | $0.75 | $0.07 | $3.75 | Same 1 January 2027 change. | |
| Gemini 3.6 Flash | $0.75 | $0.07 | $3.75 | Same 1 January 2027 change. | |
| Kimi | Kimi K2.6 | $0.95 | $0.16 | $4.00 | Context on the card: 262.1K. |
| Kimi | Kimi K2.7 Code | $0.95 | $0.19 | $4.00 | Context on the card: 262.1K. |
| Meta Muse | Muse Spark 1.3 Standard | $1.25 | $0.15 | $4.25 | Standard prompts and completions are not used to train Meta models. |
| Meta Muse | Muse Spark 1.2 Standard | $1.25 | $0.15 | $4.25 | Same Standard data-use line. |
| Meta Muse | Muse Spark 1.1 Standard | $1.25 | $0.15 | $4.25 | Same Standard data-use line. |
| Anthropic | Claude Haiku 4.5 | $1.00 | $0.10 | $5.00 | Context on the card: 1.00M. |
| xAI | Grok 4.7 | $2.00 | $0.50 | $6.00 | Below 200K tokens. At 200K input tokens or more the page's FAQ states $4 input, $1 cached input and $12 output, applied to all tokens in that request. Separate 1.1× US endpoint rate is not the global default. Context on the card: 500.0K. |
| xAI | Grok 4.6 | $2.00 | $0.50 | $6.00 | Context on the card: 500.0K. |
| Kimi | Kimi K2.7 Code HighSpeed | $1.90 | $0.38 | $8.00 | Context on the card: 262.1K. |
| OpenAI | GPT-6 Sol | $2.00 | $0.20 | $10.00 | Long-context tier applies above 272K input tokens. |
| OpenAI | GPT-4o | $2.50 | $1.25 | $10.00 | Legacy. Context on the card: 128.0K. |
| Anthropic | Claude Sonnet 5 | $2.00 | $0.20 | $10.00 | Context on the card: 1.00M. |
| Gemini 3.1 Pro Preview | $2.00 | $0.20 | $12.00 | At or below 200K prompt tokens. The FAQ states $4 input and $18 output above that threshold. | |
| Kimi | Kimi K3 | $3.00 | $0.30 | $15.00 | Context on the card: 1.05M. Cache writes have separate TTL charges. This row is standard input and cache hits only. |
| OpenAI | GPT-5.6 Sol | $4.00 | $0.40 | $20.00 | Promotional pricing documented through at least 21 November 2026. |
| Anthropic | Claude Opus 5.5 | $4.00 | $0.20 | $20.00 | Fast mode is a research preview and cannot be combined with the Batch API. Context on the card: 1.00M. |
| Anthropic | Claude Opus 5 | $5.00 | $0.50 | $25.00 | Same fast-mode limit. Context on the card: 1.00M. |
| OpenAI | GPT-6 Astra | $10.00 | $1.00 | $50.00 | Long-context tier above 272K input tokens. Batch and Flex are documented at 50% of Standard for this model. |
| Anthropic | Claude Fable 5.1 | $10.00 | $0.25 | $50.00 | Cached figure is the hit/refresh rate. Cache writes have separate 5-minute and 1-hour charges. Context on the card: 1.00M. |
| OpenAI | GPT-5.6 Cyber | $12.50 | $1.25 | $75.00 | Specialist row. No further condition on the card. |
The US-only inference control is 1.1× on the current Claude rows. Do not apply it to other providers. Grok's US endpoint uplift is a different rule, described on the Grok 4.7 row.
If the prompt is client or regulated data, a cheaper token tier is not a governance decision. Review AI governance before you send it.
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 28 September 2026.
Optional switches
Only if that row documents them: cache-hit share, batch rate, fast mode on Opus 5 and Opus 5.5, or US-only inference on current Claude rows.
Manus credits
Not on this dollar scale. No per-million token rate on this page.
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 28 September 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.
Calculator settings
Prices as of 28 Sep 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. Tokenisation varies by provider, language, code and model; use the provider tokenizer for exact deployment counts.
Verified model estimates
Last updated: · source links are shown on every row.
Results are sorted by your current per-call estimate. Provider filtering changes the catalogue only; it does not clear your selected comparison models.
Tier
Only applies to listed cache-hit rates
Meta Muse
Muse Spark 1.3 Contributor
Input
$0.10/M
Cached
$0.002/M
Output
$0.20/M
$0.000100 per call
standard rate
Data-use warning: Contributor pricing permits Meta to use prompts and completions to train future Meta models.
Verify provider price ↗Meta Muse
Muse Spark 1.2 Contributor
Input
$0.10/M
Cached
$0.002/M
Output
$0.20/M
$0.000100 per call
standard rate
Data-use warning: Contributor pricing permits Meta to use prompts and completions to train future Meta models.
Verify provider price ↗Gemini 2.5 Flash-Lite
Input
$0.10/M
Cached
$0.01/M
Output
$0.40/M
$0.000200 per call
standard rate
Text, image and video input rate. Audio input has a separate published rate.
Verify provider price ↗OpenAI
GPT-6 Luna
Input
$0.10/M
Cached
$0.01/M
Output
$0.50/M
$0.000250 per call
standard rate
A long-context rate is published, but the cited pricing documentation does not state a safe automatic threshold for this model.
Verify provider price ↗OpenAI
GPT-4o mini
Input
$0.15/M
Cached
$0.07/M
Output
$0.60/M
$0.000300 per call
standard rate
500 request128.0K context
Gemini 3.1 Flash-Lite
Input
$0.25/M
Cached
$0.03/M
Output
$1.50/M
$0.000750 per call
standard rate
Text, image and video input rate. Audio input has a separate published rate.
Verify provider price ↗Gemini 3.5 Flash-Lite
Input
$0.30/M
Cached
$0.03/M
Output
$2.50/M
$0.001250 per call
standard rate
Verify provider price ↗xAI
Grok 4.3
Input
$1.25/M
Cached
$0.20/M
Output
$2.50/M
$0.001250 per call
standard rate
500 request1.00M context
The documented Batch API discount is 20% for this model, not 50%.
Verify provider price ↗Gemini 3.8 Flash
Input
$0.75/M
Cached
$0.07/M
Output
$3.75/M
$0.001875 per call
standard rate
Published Standard price changes to $1.50/$7.50 from 1 Jan 2027.
Verify provider price ↗Gemini 3.7 Flash
Input
$0.75/M
Cached
$0.07/M
Output
$3.75/M
$0.001875 per call
standard rate
Published Standard price changes to $1.50/$7.50 from 1 Jan 2027.
Verify provider price ↗Gemini 3.6 Flash
Input
$0.75/M
Cached
$0.07/M
Output
$3.75/M
$0.001875 per call
standard rate
Published Standard price changes to $1.50/$7.50 from 1 Jan 2027.
Verify provider price ↗Kimi
Kimi K2.6
Input
$0.95/M
Cached
$0.16/M
Output
$4.00/M
$0.002000 per call
standard rate
500 request262.1K context
Kimi
Kimi K2.7 Code
Input
$0.95/M
Cached
$0.19/M
Output
$4.00/M
$0.002000 per call
standard rate
500 request262.1K context
Meta Muse
Muse Spark 1.3 Standard
Input
$1.25/M
Cached
$0.15/M
Output
$4.25/M
$0.002125 per call
standard rate
Standard-tier prompts and completions are not used to train Meta models.
Verify provider price ↗Meta Muse
Muse Spark 1.2 Standard
Input
$1.25/M
Cached
$0.15/M
Output
$4.25/M
$0.002125 per call
standard rate
Standard-tier prompts and completions are not used to train Meta models.
Verify provider price ↗Meta Muse
Muse Spark 1.1 Standard
Input
$1.25/M
Cached
$0.15/M
Output
$4.25/M
$0.002125 per call
standard rate
Standard-tier prompts and completions are not used to train Meta models.
Verify provider price ↗Anthropic
Claude Haiku 4.5
Input
$1.00/M
Cached
$0.10/M
Output
$5.00/M
$0.002500 per call
standard rate
500 request1.00M context
xAI
Grok 4.7
Input
$2.00/M
Cached
$0.50/M
Output
$6.00/M
$0.003000 per call
standard rate
500 request500.0K context
The cited xAI pricing page documents a separate 1.1× US endpoint rate.
Verify provider price ↗xAI
Grok 4.6
Input
$2.00/M
Cached
$0.50/M
Output
$6.00/M
$0.003000 per call
standard rate
500 request500.0K context
Kimi
Kimi K2.7 Code HighSpeed
Input
$1.90/M
Cached
$0.38/M
Output
$8.00/M
$0.004000 per call
standard rate
500 request262.1K context
OpenAI
GPT-6 Sol
Input
$2.00/M
Cached
$0.20/M
Output
$10.00/M
$0.005000 per call
standard rate
The documented long-context tier applies above 272K input tokens.
Verify provider price ↗OpenAI
GPT-4o
Input
$2.50/M
Cached
$1.25/M
Output
$10.00/M
$0.005000 per call
standard rate
500 request128.0K context
Anthropic
Claude Sonnet 5
Input
$2.00/M
Cached
$0.20/M
Output
$10.00/M
$0.005000 per call
standard rate
500 request1.00M context
Gemini 3.1 Pro Preview
Input
$2.00/M
Cached
$0.20/M
Output
$12.00/M
$0.006000 per call
standard rate
The published breakpoint is prompt size: 200K input tokens.
Verify provider price ↗Kimi
Kimi K3
Input
$3.00/M
Cached
$0.30/M
Output
$15.00/M
$0.007500 per call
standard rate
500 request1.05M context
K3 cache writes have separately documented TTL charges; this estimate prices standard input and reported cache hits only.
Verify provider price ↗OpenAI
GPT-5.6 Sol
Input
$4.00/M
Cached
$0.40/M
Output
$20.00/M
$0.0100 per call
standard rate
Official promotional pricing is documented through at least 21 Nov 2026; recheck before using this model after that date.
Verify provider price ↗Anthropic
Claude Opus 5.5
Input
$4.00/M
Cached
$0.20/M
Output
$20.00/M
$0.0100 per call
standard rate
500 request1.00M context
Fast mode is a first-party research preview and cannot be combined with the Batch API.
Verify provider price ↗Anthropic
Claude Opus 5
Input
$5.00/M
Cached
$0.50/M
Output
$25.00/M
$0.0125 per call
standard rate
500 request1.00M context
Fast mode is a first-party research preview and cannot be combined with the Batch API.
Verify provider price ↗OpenAI
GPT-6 Astra
Input
$10.00/M
Cached
$1.00/M
Output
$50.00/M
$0.0250 per call
standard rate
The official model page documents a >272K-input-token long-context tier; Batch and Flex are documented at 50% of Standard for this model.
Verify provider price ↗Anthropic
Claude Fable 5.1
Input
$10.00/M
Cached
$0.25/M
Output
$50.00/M
$0.0250 per call
standard rate
500 request1.00M context
The cache figure is the official cache hit/refresh rate; cache writes have separate 5-minute and 1-hour charges.
Verify provider price ↗OpenAI
GPT-5.6 Cyber
Input
$12.50/M
Cached
$1.25/M
Output
$75.00/M
$0.0375 per call
standard rate
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 28 September 2026 list price.
| Model | Input | Output | Total / call | vs lowest |
|---|---|---|---|---|
Gemini 3.8 Flash | $0.00 | $0.001875 | $0.001875 | Lowest |
Meta Muse Muse Spark 1.3 Standard | $0.00 | $0.002125 | $0.002125 | 1.1× |
xAI Grok 4.7 | $0.00 | $0.003000 | $0.003000 | 1.6× |
Kimi Kimi K3 | $0.00 | $0.007500 | $0.007500 | 4.0× |
OpenAI GPT-6 Astra | $0.00 | $0.0250 | $0.0250 | 13.3× |
Anthropic Claude Fable 5.1 | $0.00 | $0.0250 | $0.0250 | 13.3× |
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
- Paste a representative prompt and set the response length you expect.
- Compare the quality, latency and context needs of a small model shortlist; do not choose on price alone.
- Turn on cache, batch, fast mode or US inference only if your actual provider configuration uses that documented feature.
- Validate tokenisation, 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 tokens | Manus credits | |
|---|---|---|
| What the page says it meters | Prompt and response text, priced per million tokens | Agent tasks, which can combine research, browser work, code, files and model use |
| Dollar rate on this page | The table above, dated 28 September 2026 | None. Do not invent one. |
| How to try it | Provider pricing pages linked in the existing FAQ | The 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.
Input tokens are the prompt, instructions and retrieved context sent to a model. Output tokens are the response the model generates, and are often priced higher because generation is an incremental process. The exact gap varies by model, so this calculator keeps the two rates separate.
A context window is the amount of information a model can consider in a request, including the prompt and generated response. Large windows can be useful for documents, conversations and retrieval workflows, but some providers publish a different price once an input crosses a threshold. The calculator automatically applies a documented long-context tier only where a first-party source states its trigger.
The calculator estimates input tokens from text using a four-characters-per-token heuristic, then applies the selected model's published text input and output price per million tokens. The calculation is input tokens divided by one million times the input rate, plus output tokens divided by one million times the output rate. Optional cache, batch, fast-mode and US-residency choices are applied only for models with a published rule.
The price arithmetic is exact for the entered token quantities and the displayed source-date price schedule. The text estimate is only a planning approximation, not a provider tokenizer. Treat the result as a budget estimate and verify both tokenisation and the current provider page before committing production spend.
Prompt caching allows a provider to reuse an eligible prefix such as a stable system instruction, long document or tool definition. When a provider reports a cache hit, its published cached-input rate can be much lower than the regular input rate. Savings depend on eligibility, prefix reuse, TTL and actual cache hits, so the slider is an estimate rather than a guarantee.
Batch APIs process work asynchronously and may have a lower published rate than interactive API calls. This tool applies batch pricing only for models whose first-party documentation publishes a batch rate, and leaves other rows on their standard rate. Batch is not appropriate for every time-sensitive workflow and can be incompatible with provider-specific fast modes.
Long-context pricing is a published higher tier that some providers apply when a request's prompt exceeds a stated token threshold. For example, Gemini 3.1 Pro Preview and several Grok models publish a 200K input-token threshold, and xAI states that the higher rate then applies to all tokens in that request. This tool applies those rules automatically only where the threshold is explicitly documented.
The lowest standard text price in a particular selection is not automatically the best operational choice, because quality, retries, context requirements and safety controls also affect total cost. Current budget rows in this catalogue include GPT-6 Luna, Gemini Flash-Lite, Claude Haiku 4.5, Kimi K2.6 and Muse Spark Contributor, each with different terms. Compare the output quality and provider conditions before routing high-volume work.
There is no universal best-value model for complex work. A sensible evaluation compares task quality, tool support, context needs, latency, reliability and the cost of retries rather than using a price row alone. Use the calculator to build a short list, then test a representative workflow under the provider terms that apply to your account.
The selected-model table prices whatever text is in the calculator. An empty prompt has no input tokens, so the input side is $0 and the row is only the output charge for the default 500 tokens. That is not the list price. Use the standard-rate table, dated 28 September 2026, for dollars per million tokens, and verify the provider page before you commit spend.
At the 28 September 2026 source check, GPT-6 Astra is listed at $10 input and $50 output per million text tokens, GPT-6 Sol at $2/$10, and GPT-6 Luna at $0.10/$0.50. GPT-5.6 Sol is listed at $4/$20 under documented promotional pricing, while GPT-5.6 Cyber is $12.50/$75. GPT-4o remains in the catalogue as a legacy comparison at $2.50/$10.
The verified standard API schedule lists Claude Fable 5.1 at $10 input and $50 output per million tokens, Opus 5.5 at $4/$20, Opus 5 at $5/$25, Sonnet 5 at $2/$10 and Haiku 4.5 at $1/$5. Anthropic separately publishes cache reads and writes, batch prices and selected fast-mode prices. The calculator uses a cache-hit estimate where the documentation provides one.
Gemini 3.8, 3.7 and 3.6 Flash are listed at $0.75 input and $3.75 output per million text tokens through 31 December 2026, followed by a published higher 2027 schedule. Gemini 3.1 Pro Preview is $2/$12 at or below 200K prompt tokens and $4/$18 above that threshold. Google also lists separate batch, flex, priority, caching, storage and modality prices.
xAI lists Grok 4.7 at $2 input, $0.50 cached input and $6 output per million tokens for prompts below 200K tokens. At the published long-context tier of 200K tokens or more, the rates are $4 input, $1 cached input and $12 output, and xAI states the higher rate applies to all tokens in that request. A US endpoint has a separate published uplift and is not selected by this global-default estimate.
Moonshot lists Kimi K3 at $3 input, $0.30 cached input and $15 output per million tokens, with a 1,048,576-token context window. The same source lists separate cache-write charges for selected TTLs. This calculator uses the standard input and cache-hit rates rather than guessing TTL behaviour.
Muse Spark is available through the Meta Model API with Standard and Contributor tiers. The verified Standard rows are $1.25 input, $0.15 cached input and $4.25 output per million tokens. Meta publishes separate non-text product prices, but those are not mixed into this text-token calculator.
Muse Spark Contributor has a lower listed token price than Standard, including $0.10 input, $0.002 cached input and $0.20 output per million tokens for the documented Contributor rows. In exchange, Meta says it may use Contributor prompts and completions to train future Meta models. Standard prompts and completions are not used for that purpose according to the cited Meta documentation.
No stable public Manus price per million API tokens is used in this calculator. Manus is presented here as a credit-based agent product, where task complexity and workflow behaviour matter more than a direct token-price conversion. The calculator therefore keeps the Manus invitation and credit explanation separate from the token-price comparison.
Use the official Manus invitation at https://manus.im/invitation/QPAHJBYCPQ3IK?utm_source=invitation&utm_medium=social&utm_campaign=system_share to try Manus with bonus credits. The offer is for credit-based Manus agent tasks, not a published per-token API rate, and programme terms can change. No credit card is required to start under the current invitation flow.
API tokens are units a model provider uses to meter prompt and response processing. Manus credits relate to agent tasks, which can combine research, browser work, code execution, files and model usage. A direct dollars-per-million-token comparison would misrepresent the different billing models, so this page does not invent one.
Start with representative input and output tokens per request, multiply by expected calls per day and then by roughly 30 days. Test both a normal request and a large-context or cached request if those are part of the design. The bulk tab makes this comparison transparent but does not include unrelated provider charges such as storage, web search, images, audio, taxes or minimum commitments.
Provider documentation can treat thinking or reasoning tokens as output tokens, include them in another usage category, or set model-specific rules. This calculator is a text input/output planning tool and follows the provider's published text schedule shown for each model. Confirm the relevant model's current documentation before budgeting an application that exposes reasoning usage.
Not necessarily. Several providers publish separate modality prices, grounding charges, image charges, audio charges or storage fees. The main grid is deliberately limited to the verified text-token rates so that it does not imply a text estimate covers multimodal usage. Check the provider source links for the applicable modality schedule.
USD list prices are useful for a common provider comparison, but a UK budget should also account for exchange-rate exposure, VAT or other taxes where applicable, cloud or platform charges, volume discounts and contractual commitments. This page does not present a live GBP conversion. Convert and validate with finance and procurement before approval.
This verified catalogue is marked 28 September 2026 and should be checked again whenever a provider changes a schedule, preview model, promotion or region rule. The page deliberately shows the source date on cards and comparison tables. Price changes, availability and product terms can occur without notice.
Verify the primary source before committing spend: OpenAI https://platform.openai.com/docs/pricing, Anthropic https://platform.claude.com/docs/en/about-claude/pricing, Google https://ai.google.dev/gemini-api/docs/pricing, xAI https://docs.x.ai/developers/pricing, Moonshot https://platform.moonshot.ai/docs/pricing/chat, Meta https://ai.developer.meta.com/docs/pricing-rate-limits and Manus https://manus.im/invitation/QPAHJBYCPQ3IK?utm_source=invitation&utm_medium=social&utm_campaign=system_share.
First-party pricing sources
The model module centralises the source date and provider page for every displayed API row. Provider schedules, preview availability, promotions and terms can change.