What AI models cost today.
The list prices of every major AI model you can call directly, side by side, per million tokens. A token is a short piece of text, roughly three quarters of a word, and providers charge separately for what you send and what the model writes back. We recheck these numbers whenever a provider changes its prices.
Prices checked October 20, 2026Price your own workload
- current models tracked
- 19
- lowest input price per million tokens, on gpt-oss-20b (Groq)
- $0.075
- gap between the cheapest and dearest output price
- 167×
- when prices were last checked
- October 20
Output prices, highest first
What the model writes is the expensive part: on most models output costs four to five times as much as input, so long answers move the bill more than long prompts. The cheapest model on the list is marked.
US dollars per million output tokens, standard tier, list prices checked October 20, 2026.
Input prices
What you send: instructions, pasted documents and the conversation so far. Many providers also sell repeated input at a large discount when it is cached, shown in the full table below.
US dollars per million input tokens, standard tier, list prices checked October 20, 2026.
What everyday tasks cost a month
Prices per million tokens are hard to picture, so here are three common jobs priced on a spread of models. Each chart states its assumptions; change them to your own in the AI Cost Calculator.
Monthly cost of 1,500 calls with about 800 tokens in and 300 out each. List prices checked October 20, 2026; your bill may differ.
Monthly cost of 200 calls with about 7,000 tokens in and 600 out each. List prices checked October 20, 2026; your bill may differ.
Monthly cost of 60,000 calls with about 4,500 tokens in and 350 out each, 70% of the input read from cache. List prices checked October 20, 2026; your bill may differ.
Every model, every price
| Provider | Model | Tier | Input | Output | Cached input | Context |
|---|---|---|---|---|---|---|
| Anthropic | Claude Fable 5.1 | Most capable | $10 | $50 | $0.25 | 1M |
| Anthropic | Claude Opus 5.5 | Most capable | $4.00 | $20 | $0.20 | 1M |
| Anthropic | Claude Sonnet 5.5 | Balanced | $2.00 | $10 | $0.10 | 1M |
| Anthropic | Claude Haiku 5.5Up to 100K-token prompts; $0.50 / $2.50 beyond. | Fast and cheap | $0.10 | $0.50 | $0.01 | 1M |
| DeepSeek | DeepSeek V4 ProPeak-hour price (01–04 and 06–10 UTC, Monday to Friday); half price at all other times. | Most capable | $1.32 | $3.96 | $0.044 | 1M |
| DeepSeek | DeepSeek FlashPeak-hour price (01–04 and 06–10 UTC, Monday to Friday); half price at all other times. | Fast and cheap | $0.30 | $1.20 | $0.006 | 1M |
| Gemini 4 ArgonIntroductory price until 2026-11-30, then $4 / $20. | Most capable | $2.00 | $10 | $0.10 | 1.05M | |
| Gemini 3.8 FlashFree tier available. $1.50 / $7.50 from 2027-01-01. | Balanced | $0.75 | $3.75 | $0.075 | 1.05M | |
| Gemini 3.5 Flash-LiteFree tier available. | Fast and cheap | $0.30 | $2.50 | $0.03 | 1.05M | |
| Groq | gpt-oss-120b (Groq) | Balanced | $0.15 | $0.60 | $0.075 | 131K |
| Groq | gpt-oss-20b (Groq) | Fast and cheap | $0.075 | $0.30 | $0.037 | 131K |
| Mistral | Mistral Medium 3.5 | Balanced | $1.50 | $7.50 | – | 256K |
| Mistral | Mistral Large 4Preview, sale price (regular $1.36 / $4.18). | Most capable | $0.68 | $2.09 | $0.07 | 1M |
| Mistral | Mistral Small 4 | Fast and cheap | $0.15 | $0.60 | – | 256K |
| OpenAI | GPT-6 AstraAbove 272K input tokens: 2× input, 1.5× output. | Most capable | $10 | $50 | $1.00 | 1.05M |
| OpenAI | GPT-6.1 SolAbove 272K input tokens: 2× input, 1.5× output. | Balanced | $2.00 | $10 | $0.10 | 1.05M |
| OpenAI | GPT-6 Luna | Fast and cheap | $0.10 | $0.50 | $0.01 | 1.05M |
| xAI | Grok 4.72× on prompts of 200K tokens or more. | Most capable | $2.00 | $6.00 | $0.50 | 500K |
| xAI | Grok 4.32× on prompts of 200K tokens or more. | Balanced | $1.25 | $2.50 | $0.20 | 1M |
US dollars per million tokens, standard tier, list prices checked October 20, 2026. Context is the most text a model takes in at once, in tokens. Your provider's pricing page and bill are the final word.
How to read these prices
- The cheapest model is not the cheapest choice if it gets the job wrong and a person has to fix it. Test two or three candidates on your own examples in the Prompt Tester or Model Compare, then pick the cheapest one that passes.
- Caching and batch processing can cut a bill by half or more. Put the fixed part of a prompt first so it can be cached, and send work that can wait as a batch.
- Some prices change with conditions: very long prompts, set dates or time of day. Those notes sit under the model's name in the table.
- Free tiers from Google AI Studio, OpenRouter and Groq cost nothing within their limits, which is plenty for trying the tools on this site.
Want an email when a model you use changes price? Price alerts are planned for Pro. Tell us you want them.