Mistral Small 4 — API Pricing & Benchmarks
Mistral Small 4 is Mistral AI's small reasoning model, released in 2026-03. The API costs $0.15 per 1M input tokens and $0.60 per 1M output tokens ($0.26 blended at a 3:1 ratio). On the independently measured Artificial Analysis Intelligence Index it scores 19.7, ranking #57 of 68 models tracked on this site, with a median output speed of 164 tokens/sec.
Explore provider routes and additional benchmark sources for Mistral Small 4
| Provider | Mistral AI |
| Input price / 1M tokens | $0.15 |
| Output price / 1M tokens | $0.60 |
| Cached input / 1M tokens | — |
| Blended price (3:1) | $0.26 |
| Context window | — |
| Max output tokens | — |
| Knowledge cutoff | — |
| License | open weights |
| Input modalities | T+I |
| Output speed (tok/s) | 164 |
| Released | 2026-03 |
| Artificial Analysis Intelligence Index | 19.7 |
| GPQA Diamond | 76.9% |
| Humanity's Last Exam | 9.9% |
Where Mistral Small 4 fits
On a blended 3:1 basis Mistral Small 4 costs $0.26 per 1M tokens, cheaper than 89% of the 80 priced models in this index. It is built for high-volume, latency-sensitive work — classification, extraction, routing and autocomplete — where per-token price matters more than peak intelligence. Measured output speed of 164 tokens/sec is well above the index median of 104, making it suitable for interactive, user-facing use. The weights are open, so it can be self-hosted for data-residency or cost reasons instead of consumed via API.
Price history
| Period | Input $/1M | Output $/1M | Cached $/1M |
|---|---|---|---|
| 2026-07-18 → current | $0.15 | $0.60 | — |
No list-price changes recorded since tracking began. Full dataset: price-history.json (CC BY 4.0).
Alternatives at a similar price
- GPT-4o mini (OpenAI) — $0.26/1M blended, AA Index 6.7
- Solar Pro 3 (Upstage) — $0.26/1M blended, AA Index 14.5
- Solar Pro 2 (Upstage) — $0.26/1M blended, AA Index 12.5
- GLM-5.3-Flash (Zhipu AI (Z.ai)) — $0.24/1M blended, AA Index 57.5
About Mistral AI
Mistral AI is Europe's leading LLM lab, shipping a mix of open-weights models (Mistral Small, Large) and proprietary ones, plus the code-focused Codestral line. European data-residency requirements often shortlist Mistral by default.