Mistral Small 3

Mistral AI has introduced Mistral Small 3, a latency-optimized 24-billion-parameter model released under the Apache 2.0 license. Designed for low latency, it is competitive with larger models like Llama 3.3 70B and is presented as an open-source alternative to proprietary models such as GPT4o-mini. The model shows strong performance on benchmarks for code, math, and general knowledge, achieving over 81% accuracy on MMLU while being significantly faster than competitors. Ideal use cases include fast conversational assistance, function calling, and local inference, with the model now available on platforms like Hugging Face, Ollama, and la Plateforme.