One agent session
1M input tokens and 100K output tokens, no caching.
- Input1M x $0.15 = $0.15
- Output0.1M x $0.60 = $0.06
$0.21per session
Mistral AI - Open weights
Mistral Small 4 pricing and specs
Mistral Small 4 is an open-weights model from Mistral AI that costs $0.15 per 1M input tokens and $0.60 per 1M output tokens on the first-party API, with a 256K-token context window.
Last verified from docs.mistral.ai. All sources
- Input / 1M
- $0.15
- Output / 1M
- $0.60
- Context
- 256K tokens
- Weights
- Open
Pricing per 1M tokens
| Item | Price / 1M tokens |
|---|---|
| Input | $0.15 |
| Output | $0.60 |
| Cache read | $0.015 |
| Batch inputBatch requests | $0.075 |
| Batch outputBatch requests | $0.30 |
Note: Batch prices come from the pricing page's Batch mode, which applies a factor of 0.5 to standard prices. Priority Tier adds 75% and regional inference adds 10%. The docs give the context as '256k' (the changelog also says 256k), and the Vercel AI Gateway lists 262144. No primary page states a max output. The model card links its weights to huggingface.co/mistralai/Mistral-Small-4-2603, which returned 404. The official mistralai repo is Mistral-Small-4-119B-2603 (license:apache-2.0), confirmed through the Hugging Face API. The gateway id mistral/mistral-small is a generic alias. It is mapped to Small 4 only because its price ($0.15/$0.6, $0.015 cache read) and its 262144 context match this model. The model card is dated March 16, 2026; release date uses the changelog date (March 15).
Specs
- Context window
- 256,000 tokens
- Release date
- Provider
- Mistral AI
- API model id
mistral-small-2603- Vercel AI Gateway id
mistral/mistral-small- Weights
- Open weights
- License
- Apache 2.0
- Weights download
- huggingface.co
What Mistral Small 4 costs in practice
Worked from the standard list rates above, with the math shown so you can swap in your own token counts.
A chat app
10,000 requests a day at 2K input and 500 output tokens each.
- Input20M x $0.15 = $3.00
- Output5M x $0.60 = $3.00
$6.00per day
$180.00 per 30-day month
The same session at a 90% cache hit
90% of the input is read from the prompt cache. The one-time cache write premium is not included.
- Cached input0.9M x $0.015 = $0.0135
- Uncached input0.1M x $0.15 = $0.015
- Output0.1M x $0.60 = $0.06
$0.0885per session
Compare with
| Model | Input / 1M | Output / 1M |
|---|---|---|
| Mistral Small 4 (this page) | $0.15 | $0.60 |
| Command R 08-2024Cohere | $0.15 | $0.60 |
| GPT-4o MiniOpenAI | $0.15 | $0.60 |
| Hy3Tencent | $0.132 | $0.528 |
| Claude Haiku 5.5Anthropic | $0.10 | $0.50 |
| Codestral 25.08Mistral AI | $0.30 | $0.90 |
| Ministral 3 14BMistral AI | $0.20 | $0.20 |
| Ministral 3 8BMistral AI | $0.15 | $0.15 |
| Ministral 3 3BMistral AI | $0.10 | $0.10 |
| Mistral Large 3Mistral AI | $0.50 | $1.50 |
| Mistral Large 4Mistral AI | $1.36 | $4.18 |
Compare
Side-by-side pages with both models' prices, specs, cost examples and like-for-like benchmark rows.
- Mistral Small 4 vs Claude Haiku 5.5Similar price, same tier
- Mistral Small 4 vs GPT-6 LunaSimilar price, same tier
- Mistral Small 4 vs DeepSeek V4.1 FlashSimilar price, same tier
- Mistral Small 4 vs GLM-5.3-FlashSimilar price, same tier
- Mistral Small 4 vs Qwen3.8-FlashSimilar price, same tier
- Mistral Small 4 vs Gemini 3.5 Flash-LiteSimilar price, same tier
- Mistral Small 4 vs MiniMax-M3Similar price, same tier
Related reading
- Mistral Large 4: Specs, Pricing, Benchmarks and Weights DateMistral Large 4 is a 1T-parameter multimodal MoE with 52B active, in public preview at $0.68/$2.09 per million tokens, with open weights due end of October.
- Mistral Shieldstral: A 3B Open-Weight Policy-Adaptive Moderation Model That Beats Models 7x Its SizeMistral Shieldstral is a 3B Apache 2.0 multimodal moderation model that takes your policy as a plain-language question at inference and runs on one 16GB GPU.
- Mistral Releases Robostral Navigate: An 8B Robotics Navigation ModelMistral Robostral Navigate is an 8B robotics model that navigates with only a camera and natural language commands. How it works and what the benchmarks mean.
- Leanstral 1.5: Mistral's Open Theorem-Proving Model Hits 100% on miniF2FLeanstral 1.5 is Mistral's open Apache-2.0 model for Lean 4 theorem proving. The benchmark claims, the HN pushback on its bug-finding pitch, and how to use it.
- Mistral OCR 4 and Unlimited OCR Make Document Parsing an Agent Runtime ChoiceMistral OCR 4 and Baidu's Unlimited OCR turn document parsing into a runtime choice for agents - cost per page, layout, confidence scores and self-hosting.
- Codex Custom Model Providers: GLM-5.3, Mistral and OllamaCodex only speaks the Responses API now. Tested config.toml blocks for GLM-5.3, Mistral Large 4 through a gateway, local Ollama and an internal proxy.
FAQ
How much does Mistral Small 4 cost per million tokens?
Mistral Small 4 costs $0.15 per 1M input tokens and $0.60 per 1M output tokens on Mistral AI's first-party API, per docs.mistral.ai on October 17, 2026. Per 1M tokens, cache reads cost $0.015, and the batch rate is $0.075 input and $0.30 output.
What is Mistral Small 4's context window?
Mistral Small 4 has a 256,000-token context window.
Are Mistral Small 4's weights open?
Yes. Mistral Small 4's weights are published at https://huggingface.co/mistralai/Mistral-Small-4-119B-2603 under the Apache 2.0.
When was Mistral Small 4 released?
Mistral AI released Mistral Small 4 on March 15, 2026.
Sources
Last verified from docs.mistral.ai.
Newsletter
Get model price changes as they ship
One email per week. New model releases, price cuts and what they mean for your bill.