Mistral Small 3.1

Mistral Small 3.1 extends its predecessor's text capabilities by adding robust multimodal understanding, enabling the model to reason about images alongside written content. Its design emphasizes versatility across programming, mathematical reasoning, document comprehension, and conversational tasks, targeting developers who need strong AI assistance without the resource demands of larger systems. The expanded context window allows it to process and maintain coherence across lengthy documents and extended exchanges.

The model demonstrates competitive benchmark performance against similarly-sized offerings, matching or surpassing results on evaluations like MMLU, GPQA, and HumanEval according to Mistral's testing. Its permissive Apache 2.0 licensing differentiates it from many commercial alternatives, making it attractive for teams building production applications with flexibility for commercial deployment. Inference speeds around 150 tokens per second make it practical for real-time use cases, positioning it as a capable foundation for a wide range of AI-powered workflows.