mistralai/Mistral-Small-4-119B-2603
119B • Updated • 1.87k • 217
A state-of-the-art model, open-weight, with a granular Mixture-of-Experts architecture that fuses instruct, reasoning and agentic skills.
Note The FP8 checkpoint to ensure best accuracy.
Note The NVFP4 checkpoint to improve your throughput and reduce memory usage. Expect lower performance on long context.
Note To increase your throughput, this eagle head brings speculative decoding to Mistral Small 4.