Mistral Small 3.1 24B

Mistral Small 3.1 24B Instruct is a multimodal 24-billion-parameter model that brings vision understanding and extended context handling to Mistral's compact Small family, building on the lineage from Mistral Small 3 2501. It processes both text and image inputs while producing text output, supporting over 22 languages including English, French, Spanish, German, Chinese, Japanese, and many others. The model achieves around 150 tokens per second in inference, making it fast enough for conversational assistants and interactive applications. It outperforms comparable models like Gemma 3 and GPT-4o Mini, and can run on a single RTX 4090 GPU or a Mac with 32GB RAM, enabling efficient local inference for privacy-sensitive deployments.

This instruction-finetuned variant derives from the Mistral-Small-3.1-24B-Base-2503 base model, adding state-of-the-art vision capabilities alongside the extended context handling without sacrificing performance on standard language tasks. The model is available under Apache 2.0 licensing, making it fully open for commercial use and fine-tuning. Organizations can specialize it for domains like legal advice, medical diagnostics, or technical support through targeted fine-tuning, creating accurate subject matter experts. Its lightweight footprint combined with strong performance in text, code, math, and vision tasks makes it particularly practical for teams that need advanced AI capabilities without relying on cloud services or enterprise budgets.