free tool · runs in your browser
AI model price calculator
· every price checked 18 Oct 2026 against the official pricing page · see sources
Compare what one request, one day and one month of API usage costs on Claude, GPT, Gemini, DeepSeek and Mistral coding models. Enter your token counts, how much of the input is cached and how many requests you send a day. Cheapest model first.
| Model | Per request | Per day | Per month |
|---|---|---|---|
| The calculator needs JavaScript. The prices it uses are listed below. | |||
Standard API rates in USD, no batch discounts, fast modes, data residency uplifts or tool fees. Cache writes are not included (see the caching caveat). Month = 30 days.
How to estimate your tokens
- Rule of thumb: about 4 characters or 0.75 words of English per token. Code, JSON and non-English text use more tokens per character.
- Agents resend everything: each turn's input is the system prompt, tool definitions, the files it read and the whole conversation so far. A coding agent turn is often 20K to 100K input tokens even when the reply is short.
- Thinking counts as output: reasoning tokens are billed at the output rate on these models, so set output higher than the visible answer.
- Tokenizers differ: Anthropic says Claude 4.7 and later produce about 30% more tokens for the same text than older Claude models. The same prompt is not the same token count on every model.
- Measure it: run a few real requests and read the usage field in the API response, then put those numbers in here.
The caching caveat
The cached share is the part of each request's input read from the provider's prompt cache. Cache hits are cheap (from 0.05x to 0.1x of the input price on most models here, and about 2% on DeepSeek), so a stable system prompt or repo context can cut input cost a lot. But:
- Writes cost extra. Anthropic charges 1.25x input for 5-minute cache writes and 2x for 1-hour writes. OpenAI lists cache write prices for its newest models. This calculator leaves writes out, so heavy cache churn costs more than shown.
- Hits are not guaranteed. Caches expire (Anthropic's default lasts 5 minutes), and Gemini 4 Argon only has implicit caching. The first request always pays full price.
- Long prompts change tiers. On Claude Haiku 5.5 the 100K threshold counts cached tokens too, so a big cached context can still push the whole request onto the higher rate.
Prices used checked 18 Oct 2026
USD per 1M tokens, standard tier. Where a model has two prices, the second applies to long prompts (over 100K input tokens on Haiku 5.5, over 272K on GPT models).
| Model | Input | Cached input | Output | Notes |
|---|---|---|---|---|
| Claude Haiku 5.5 Anthropic · claude-haiku-5-5 | $0.10 / $0.50 | $0.01 / $0.05 | $0.50 / $2.50 | Prompts over 100K tokens (all input, cache reads and writes included) pay $0.50 / $2.50 for the whole request. checked 18 Oct 2026 · source |
| Claude Sonnet 5.5 Anthropic · claude-sonnet-5-5 | $2.00 | $0.10 | $10.00 | Cache hits are 0.05x input. Same rate across the 1M context. checked 18 Oct 2026 · source |
| Claude Opus 5.5 Anthropic · claude-opus-5-5 | $4.00 | $0.20 | $20.00 | Cache hits are 0.05x input. Same rate across the 1M context. checked 18 Oct 2026 · source |
| Gemini 4 Argon (through 30 Nov) Google · gemini-4-argon | $2.00 | $0.10 | $10.00 | Intro price until 30 Nov 2026. Output includes thinking tokens. Implicit caching only. checked 18 Oct 2026 · source |
| Gemini 4 Argon (from 1 Dec) Google · gemini-4-argon | $4.00 | $0.20 | $20.00 | Price doubles on 1 Dec 2026. Output includes thinking tokens. Implicit caching only. checked 18 Oct 2026 · source |
| GPT-6 Astra OpenAI · gpt-6-astra | $10.00 / $20.00 | $1.00 / $2.00 | $50.00 / $75.00 | Over 272K input tokens: long-context rates. checked 18 Oct 2026 · source |
| GPT-6.1 Sol OpenAI · gpt-6.1-sol | $2.00 / $4.00 | $0.10 / $0.20 | $10.00 / $15.00 | Over 272K input tokens: long-context rates. checked 18 Oct 2026 · source |
| GPT-6 Sol OpenAI · gpt-6-sol | $2.00 / $4.00 | $0.20 / $0.40 | $10.00 / $15.00 | Over 272K input tokens: long-context rates. checked 18 Oct 2026 · source |
| GPT-6 Luna OpenAI · gpt-6-luna | $0.10 / $0.20 | $0.01 / $0.02 | $0.50 / $0.75 | Over 272K input tokens: long-context rates. checked 18 Oct 2026 · source |
| GPT-5.6 Sol OpenAI · gpt-5.6-sol | $4.00 / $8.00 | $0.40 / $0.80 | $20.00 / $30.00 | Promotional price, at least through 21 Nov 2026. Over 272K input tokens: long-context rates. checked 18 Oct 2026 · source |
| GPT-5.6 Terra OpenAI · gpt-5.6-terra | $2.00 / $4.00 | $0.20 / $0.40 | $12.00 / $18.00 | Over 272K input tokens: long-context rates. checked 18 Oct 2026 · source |
| GPT-5.6 Luna OpenAI · gpt-5.6-luna | $0.20 / $0.40 | $0.02 / $0.04 | $1.20 / $1.80 | Over 272K input tokens: long-context rates. checked 18 Oct 2026 · source |
| GPT-5.3 Codex OpenAI · gpt-5.3-codex | $1.75 | $0.175 | $14.00 | Codex model. checked 18 Oct 2026 · source |
| DeepSeek V4.1 Flash (off-peak) DeepSeek · deepseek-flash | $0.15 | $0.003 | $0.60 | Off-peak: all hours outside the peak windows, plus weekends and Chinese public holidays. checked 18 Oct 2026 · source |
| DeepSeek V4.1 Flash (peak) DeepSeek · deepseek-flash | $0.30 | $0.006 | $1.20 | Peak: 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday. Twice the off-peak price. checked 18 Oct 2026 · source |
| Mistral Large 4 Mistral · mistral-large-4 | $0.68 | $0.07 | $2.09 | Public preview, sale price (list $1.36 / $4.18, cached $0.14). End date not stated. checked 18 Oct 2026 · source |
OpenAI's pricing page also lists older GPT-5.x, GPT-4.x and o-series models. We include the current flagship family and GPT-5.3 Codex. Spotted a price change? Email commitweekly@mail.grokbot.com.
FAQ
How do I estimate how many tokens my requests use?
A rough rule for English text is about 4 characters or 0.75 words per token. Code and JSON usually take more tokens per character. For agents, count everything sent each turn: system prompt, tool definitions, file contents and the whole conversation so far. The most accurate number is the usage field in a real API response, or the provider's token counting endpoint.
What does cached share mean?
It is the share of each request's input tokens that the provider reads from its prompt cache instead of processing again. Cache hits are billed at the cached input rate, which is 5% to 10% of the normal input price for most models here. A long, stable system prompt or codebase context that you resend on every turn is the typical cached part.
Does the calculator include cache write costs?
No. Anthropic charges 1.25x the input price to write to the 5-minute cache and 2x for the 1-hour cache, and OpenAI lists separate cache write prices for its newest models. Gemini 4 Argon only has implicit caching, so hits are not guaranteed. Treat the cached result as a best case and check your real usage.
Why do some models change price with prompt length?
Claude Haiku 5.5 charges a higher rate for the whole request once the prompt is over 100K tokens, and OpenAI's flagship GPT models switch to long-context rates above 272K input tokens. The calculator picks the right tier from your input token count.
How often are the prices updated?
Every price links to the official pricing page with the date it was last checked. Prices change often: Gemini 4 Argon doubles on 1 Dec 2026, DeepSeek charges double at peak hours, and some prices are promotions. Check the source before you commit to a budget.
Picking a model for a real job? Read our notes on Claude Haiku 5.5 (the cheap subagent) and Gemini 4 Argon (the frontier planner).
// subscribe
Get new tools and price changes every Tuesday free
One email a week with the AI tools worth trying. Subscribing happens on Beehiiv, our newsletter platform.
No form? or subscribe on Beehiiv → · free · Tuesdays · unsubscribe anytime