Mistral Small 3.1

Mistral AI · 1 year ago

Mistral released Small 3.1, an open-source model with improved text performance, multimodal capabilities, and a context window of 128,000 tokens that runs on consumer hardware like an RTX 4090 or 32GB Mac. The model achieves inference speeds of 150 tokens per second and outperforms comparable models including Gemma 3 and GPT-4o Mini across multiple benchmarks. The model is available immediately via Hugging Face and Mistral's API, with deployments on Google Cloud, NVIDIA NIM, and Microsoft Azure coming in the following weeks.