French AI startup Mistral announced the release of Mistral Small 3.1, its new lightweight model, on Monday. It’s a state-of-the-art multimodal, multilingual, open-source model available under an Apache licence.
The new model builds on Mistral Small 3, with improved text performance, multimodal understanding, and an expanded context window of up to 128k tokens. Mistral claims that the model outperforms comparable models like Gemma 3 and GPT-4o mini while delivering inference speeds of 150 tokens per second.
According to the official performance chart in the announcement post, Mistral Small 3.1 surpasses other leading small proprietary models in various applications involving text, multimodal inputs, and managing long contexts.
The blog post stated, “Mistral Small 3.1 is a versatile model designed to handle a wide range of generative AI tasks, including instruction following, conversational assistance, image understanding, and function calling. It provides a solid foundation for both enterprise and consumer-grade AI applications.”
The announcement also highlights that the model is lightweight and can run on a single RTX 4090 or a Mac with 32 GB RAM. Moreover, it is ideal for virtual assistants and applications where quick, accurate responses are needed.
Mistral has clarified that the model can be fine-tuned to specialise in specific domains and also made the pre-trained base model available along with the release. The company mentioned that it can be used across various enterprise and consumer applications that require multimodal understanding.
The model can be downloaded through Hugging Face. One can also try it via API on Mistral AI’s developer playground, La Plateforme; it is also available on Google Cloud Vertex AI.
It will be available on NVIDIA NIM and Microsoft Azure AI Foundry in the coming weeks.
