Today we announce Mistral Small 3.1: the best model in its weight class.
Building on Mistral Small 3, this new model comes with improved text performance, multimodal understanding, and an expanded context window of up to 128k tokens. The model outperforms comparable models like Gemma 3 and GPT-4o Mini, while delivering inference speeds of 150 tokens per second. (View Highlight)
Modern AI applications demand a blend of capabilities—handling text, understanding multimodal inputs, supporting multiple languages, and managing long contexts—with low latency and cost efficiency. As shown below, Mistral Small 3.1 is the first open source model that not only meets, but in fact surpasses, the performance of leading small proprietary models across all these dimensions. (View Highlight)
• Lightweight: Mistral Small 3.1 can run on a single RTX 4090 or a Mac with 32GB RAM. This makes it a great fit for on-device use cases.
• Fast-response conversational assistance: Ideal for virtual assistants and other applications where quick, accurate responses are essential.
• Low-latency function calling: Capable of rapid function execution within automated or agentic workflows
• Fine-tuning for specialized domains: Mistral Small 3.1 can be fine-tuned to specialize in specific domains, creating accurate subject matter experts. This is particularly useful in fields like legal advice, medical diagnostics, and technical support.
• Foundation for advanced reasoning: We continue to be impressed by how the community builds on top of open Mistral models. Just in the last few weeks, we have seen several excellent reasoning models built on Mistral Small 3, such as the DeepHermes 24B by Nous Research. To that end, we are releasing both base and instruct checkpoints for Mistral Small 3.1 to enable further downstream customization of the model. (View Highlight)
