Mistral AI's Small 4 lets an application change reasoning effort without switching models. Its reasoning_effort setting selects either fast, direct responses or more extensive reasoning for harder requests. The same model accepts text and images and is designed for chat, coding and reasoning tasks. The March 2026 announcement describes the two modes.
That gives developers one model interface for requests with different response-time needs. The release also makes the model available under Apache 2.0 for customization and deployment. Its useful unification is the ability to vary behavior within that shared model, while choosing the reasoning setting for each task.
Terms in this piece · Glossary
open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
