Overview
Mistral Small 4 replaces separate instruct, reasoning and coding models with a single hybrid. It uses 128 experts with 4 active, so about 6.5B parameters run per token, and reasoning effort can be set per request. Mistral reports 40% lower end-to-end completion time than Mistral Small 3 in a latency-optimized setup.
Highlights
- 119B total, 6.5B active parameters
- Instant and reasoning modes in one model
- Text and image input
- 256K-token context window
- EAGLE speculative-decoding head and NVFP4 checkpoint available
Typical uses
General assistants, agents with function calling, coding and document analysis on a single server.
Things to consider
- Despite the “Small” name, all 119B parameters must fit in memory.
- Speed figures are Mistral’s own measurements.
Specifications are taken from the model card on Hugging Face and the provider’s announcement. Benchmark results depend on evaluation settings; see how AI benchmarks work.