People evaluating a smaller Mistral vision model instead of the previously listed 24B model.
The documented release supports image understanding, multiple languages, system prompts, function calling and JSON output.
The catalog names Ollama 0.13.1 as the compatibility requirement; its pre-release wording may be historical, so use a current compatible runtime rather than assuming an old prerelease is required today. The advertised 256K context is not a promise of feasible context on every device. Tools require external execution.
No hosted subscription or token fee for local inference within the license’s permitted uses. Hardware, electricity and setup are separate. Apache 2.0 is documented for this release.
Install a current compatible Ollama version, review the exact model license, then run: ollama run ministral-3:8b. API means the local runtime endpoint, not a hosted service included with the weights.
Tags
© Chagto 2026. All rights reserved