Mistral Small 3
Mistral AI
Mistral released Small 3, a 24-billion-parameter open-source model designed for low-latency inference and competitive with much larger proprietary models. The model achieves 81% accuracy on MMLU benchmarks while running at 150 tokens per second, and is more than 3 times faster than Llama 3.3 70B on identical hardware. The release under Apache 2.0 license enables local deployment on consumer hardware and serves as a base for customization by the open-source community.
