Text and chat · Mistral AI
Mistral Small
A compact 24B model from Mistral AI, in private mode: very cheap, reads images and calls tools.
Private Low costWhat sets it apart
Mistral Small is a compact dense 24B model under Apache 2.0, tuned for low latency and clean function calling. It reads images and costs a fraction of larger models, but it has no reasoning mode, so hard problems are better sent elsewhere.
- Dense 24B under Apache 2.0
- Latency-first design, reads images
- 3.2 cut repetitive loops, firmer tool calls
Strengths and limits
Where it shines
- One of the cheapest models in the catalog
- Reads images, tool calling, 256K context
Think twice when
- No reasoning mode: weaker on hard problems
- 24B parameters, 256K context
- Open weights
Use cases
Quick extraction from photos
Read this photo of a restaurant receipt and list each item with its price as a table, then give the total.
It accepts images and is among the cheapest models here.
Tool-driven workflows
Look up the three most recent notes about the Lyon project and draft a status update from them.
Version 3.2 made function calling more robust.
Against its rivals
- GemmaGemma is the closest open rival that also reads images, from a newer generation.#71 of 90 on Artificial Analysis Intelligence Index (17 points)
- Llama and Hermes 3Llama 3.3 70B costs more and is text only, but it is a larger model.#86 of 90 on Artificial Analysis Intelligence Index (8 points)
- Qwen open modelsQwen open models add reasoning where Mistral Small has none.#23 of 90 on Artificial Analysis Intelligence Index (40 points)
Version history
Each version, dated, with what it changed over the one before.
Hollow points are earlier versions no longer offered in Sub Rosa.
See the numbers
| Version | Released | points |
|---|---|---|
| Mistral Small 3.2 | 2025-06-20 | 8 |
| Mistral Small 4 | 2026-03-16 | 11 |
- Mistral Small 3Earlier, not in Sub Rosa
- 24B latency-optimized model, Apache 2.0, over 3x faster than Llama 3.3 70B on the same hardware, per Mistral
- Mistral Small 3.1Earlier, not in Sub Rosa
- Added image understanding
- Context extended to 128K tokens
- Mistral Small 3.2Latest
- Better at following precise instructions: 84.78% vs 82.75% on the card’s test
- Fewer infinite or repetitive generations, a more robust function calling template
- Mistral Small 4Earlier, not in Sub Rosa
- Mixture of experts with 119B total and 6.5B active parameters, 256K context, reasoning variant
Benchmarks
How to read these scores: what a benchmark tells you.
Artificial Analysis Intelligence Index: #87 of 90 models measured in the catalog. One score averaging about ten independent tests of reasoning, knowledge, math, coding and agent work, all run the same way for every model. Higher is better. Each model is listed at its highest reasoning effort. Two or three points apart is a close call; ten points is a different class of model.
See the numbers
| Model | points |
|---|---|
| 1. Claude Opus 5.5 | 58 |
| 2. Claude Sonnet 5.5 | 56 |
| 3. Claude Fable 5.1 | 53 |
| 4. GPT-6 Astra | 53 |
| 5. Gemini 4 Argon | 53 |
| 6. GPT-6.1 Sol | 52 |
| 7. Muse Wonder | 52 |
| 8. Claude Opus 5 | 51 |
| 9. Claude Fable 5 | 50 |
| 10. GPT-6 Sol | 48 |
| 87. Mistral Small 3.2 24B Instruct | 8 |
The model in this family
| Model | Use | Privacy | Price |
|---|---|---|---|
Mistral Small 3.2 24B Instructmistral-small-3-2-24b-instruct | Text | Private | $0.112 per million tokens written$0.0000787 a page written |
Specifications
| Specification | Mistral Small 3.2 |
|---|---|
| Released | June 20, 2025 |
| Parameters | 24B |
| Weights | Open, Apache 2.0 |
| Accepts | text, images |
| Produces | text |
| Context | 256K tokens |
| Longest answer | 16K tokens |
| Abilities | Reads images, Uses Sub Rosa’s tools |
| Price | $0.112 per million tokens written |
| Privacy | Private |
Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.
Sources
Everything on this page comes from these pages, read on October 16, 2026 or as dated on each score.