Mistral Small 3.1
Mistral Small 3.1 adds visual understanding to the Small 3 line. Its 24B model accepts text and images and increases the advertised context from 32K to 128K.
Vision and long documents in one release
The card describes image analysis, long-document understanding, function calling and multilingual text. Base and Instruct checkpoints are both available under Apache 2.0. The instruction checkpoint is fine-tuned from the matching 3.1 base model, not the January text-only base.Mistral Small 3.1 official model card
Local inference conditions
The card’s single-RTX-4090 and 32GB-Mac examples require quantized weights. It also publishes the base checkpoint for further training and an instruction checkpoint for chat use. These are different training variants; a quantized copy of either remains a format of that variant.Mistral Small 3.1 official model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | Downloadable |
| Model identifiers | Mistral-Small-3.1-24B-Base-2503Mistral-Small-3.1-24B-Instruct-2503mistral-small-2503 |
| Context window | 128,000 tokens The model catalog advertises 128k context for this version. This is not a measured recall limit. Mistral Small 3.1 model and lifecycle record |
| Weight license | Apache 2.0Mistral Small 3.1 official model card |
| Access and lifecycle | Hosted API retired on 30 November 2025. Downloadable weights remain listed.Mistral Small 3.1 model and lifecycle recordMistral Small 3.1 official model card |
| Parameters | 24 billionMistral Small 3.1 official model card |