DATA
AI prices: what the API models cost
This page lists the AI prices of 46 current models via the API, in USD per 1 million tokens: Anthropic, OpenAI, Google, Mistral and open models at the model provider and at a hoster. The most expensive models, gpt-5.5-pro and gpt-5.4-pro, cost 1,800 times as much per output token as the cheapest, Ministral 3 3B.
The list is for CIOs and tech leads who want to buy, switch or run a model themselves and need to know the cost beforehand. All data is available as JSON and CSV.
46 current models · version v2026.10.24.1 · retrieved between 17 October 2026 and 23 October 2026
AI prices by model and provider
The table shows list prices without volume and batch discounts and without taxes, sorted by output price within each provider. Promotional prices are noted in the row, and the date of retrieval is shown in each provider row.
| Model | Input (USD/1M tokens) | Output (USD/1M tokens) | Cache read (USD/1M tokens) | Context window |
|---|---|---|---|---|
| Anthropicas of 17 October 2026 | ||||
| Claude Fable 5.1 | 10.00 | 50.00 | 0.25 | 1M |
| Claude Mythos 5.1limited availability | 10.00 | 50.00 | 0.25 | 1M |
| Claude Opus 5.5 | 4.00 | 20.00 | 0.20 | 1M |
| Claude Sonnet 5.5 | 2.00 | 10.00 | 0.10 | 1M |
| Claude Haiku 5.5 | 0.10 | 0.50 | 0.01 | 1M |
| OpenAIas of 23 October 2026 | ||||
| gpt-5.4-pro | 30.00 | 180.00 | – | 1.05M |
| gpt-5.5-pro | 30.00 | 180.00 | – | 1.05M |
| gpt-5.2-pro | 21.00 | 168.00 | – | 400k |
| gpt-6-astra | 10.00 | 50.00 | 1.00 | 1.05M |
| gpt-5.5 | 5.00 | 30.00 | 0.50 | 1.05M |
| gpt-5.6-solPromotional price at least through 21 Nov 2026, no follow-on rate published | 4.00 | 20.00 | 0.40 | 1.05M |
| gpt-5.4 | 2.50 | 15.00 | 0.25 | 1.05M |
| gpt-5.2 | 1.75 | 14.00 | 0.175 | 400k |
| gpt-5.6-terra | 2.00 | 12.00 | 0.20 | 1.05M |
| gpt-5.1 | 1.25 | 10.00 | 0.125 | 400k |
| gpt-6-sol | 2.00 | 10.00 | 0.20 | 1.05M |
| gpt-6.1-sol | 2.00 | 10.00 | 0.10 | 1.05M |
| gpt-5.4-mini | 0.75 | 4.50 | 0.075 | 400k |
| gpt-5.4-nano | 0.20 | 1.25 | 0.02 | 400k |
| gpt-5.6-luna | 0.20 | 1.20 | 0.02 | 1.05M |
| gpt-6-luna | 0.10 | 0.50 | 0.01 | 1.05M |
| Googleas of 22 October 2026 | ||||
| Gemini 3.1 Pro Preview | 2.00 | 12.00 | 0.20 | 1.05M |
| Gemini 3.6 FlashPromotional price through 31 Dec 2026, then 1.50 / 7.50 USD | 0.75 | 3.75 | 0.075 | 1.05M |
| Gemini 3.8 FlashPromotional price through 31 Dec 2026, then 1.50 / 7.50 USD | 0.75 | 3.75 | 0.075 | 1.05M |
| Gemini 3.5 Flash-Lite | 0.30 | 2.50 | 0.03 | 1.05M |
| Gemini 3.1 Flash-Liteshutdown announced for 7 May 2027 | 0.25 | 1.50 | 0.025 | 1.05M |
| Mistralas of 23 October 2026 | ||||
| Mistral Medium 3.5 | 1.50 | 7.50 | 0.15 | 256k |
| Mistral Large 4 | 1.36 | 4.18 | 0.14 | 1M |
| Mistral Large 3 | 0.50 | 1.50 | 0.05 | 256k |
| Codestral | 0.30 | 0.90 | 0.03 | 256k |
| Mistral Small 4 | 0.15 | 0.60 | 0.015 | 256k |
| Ministral 3 14B | 0.20 | 0.20 | 0.02 | 128k |
| Ministral 3 8B | 0.15 | 0.15 | 0.015 | 128k |
| Ministral 3 3B | 0.10 | 0.10 | 0.01 | 128k |
| Open weights, at the model provideras of 22 October 2026 | ||||
| DeepSeek V4 Pro | 1.32 | 3.96 | 0.044 | 1M |
| DeepSeek V4.1 Flash | 0.30 | 1.20 | 0.006 | 1M |
| Open weights, at a hosteras of 23 October 2026 | ||||
| Kimi K3 (Together)hosted at Together AI | 3.00 | 15.00 | 0.30 | 1M |
| Qwen3.8-2.4T-A95B (Together)hosted at Together AI | 2.00 | 6.00 | 0.25 | 1M |
| GLM-5.3 (Together)hosted at Together AI | 1.40 | 4.40 | 0.26 | 1M |
| Z.ai GLM 5.3 (Mistral)hosted at Mistral | 1.40 | 4.40 | 0.14 | 1M |
| DeepSeek V4 Pro (Together)hosted at Together AI | 1.32 | 3.96 | 0.13 | 1M |
| Qwen3.7-Plus (Together)hosted at Together AI | 0.32 | 1.28 | – | 262k |
| GLM-5.3-Flash (Together)hosted at Together AI | 0.15 | 0.50 | 0.03 | 1M |
| Z.ai GLM 5.3 Flash (Mistral)hosted at Mistral | 0.15 | 0.50 | 0.03 | 1M |
| Qwen3.8 Flash (Together)hosted at Together AI | 0.15 | 0.47 | – | 1M |
| DeepSeek V4 Flash (Together)hosted at Together AI | 0.14 | 0.28 | 0.03 | 1M |
How I read the AI prices
- For the median of the 46 models, an output token costs 5 times as much as an input token. For tasks with long answers the output price usually drives the cost, for long documents as input the input price.
- Cache reads lower the cost of repeated input, such as long system prompts or documents sent with every request.
- The list price applies to standard requests without surcharges. Batch discounts, data residency surcharges and long-context prices are only in the download.
What a task ends up costing is worked out by the AI cost calculator using the same data. Why prices are so far apart is set out in the article AI pricing models 2026: the AI price ladder.
Download AI prices as JSON and CSV
Both files contain all 101 models of the data release, including deprecated and retired ones, each with status and date of retrieval. Prices in USD per 1M tokens, dot as decimal separator.
- ki-preise.csv: one row per model, comma-separated, notes in German and English. Opens in Excel, Numbers and Google Sheets.
- ki-preise.json: the same models with all fields, plus version and date per provider. For scripts and your own calculators.
An example of the unit: Ministral 3 3B costs USD 0.10 per 1M output tokens, gpt-5.5-pro USD 180.00.
When each provider's AI prices were retrieved
The prices come from the providers' pricing pages; the date of retrieval per provider is in the table. The data release is version v2026.10.24.1.
| Provider | Retrieved |
|---|---|
| Anthropic | 17 October 2026 |
| OpenAI | 23 October 2026 |
| 22 October 2026 | |
| Mistral | 23 October 2026 |
| Open weights, at the model provider | 22 October 2026 |
| Open weights, at a hoster | 23 October 2026 |
The term behind it is explained under AI price ladder in the terms.