Mistral AI

Two products under one name — read the model card

"Mistral" is two things at once: a hosted API (La Plateforme, proprietary models like Mistral Large) and a set of open-weights models (Mistral 7B, Mixtral) you can download and run anywhere. Their licenses are not the same. Some open releases are Apache 2.0 — fully commercial-friendly — while others ship under Mistral's own research or commercial license. The model card, not the brand, tells you what you may do.

Practical implications

• For the API: inputs are not used for training by default; commercial use and output ownership are straightforward.
• For open weights: if a model is Apache 2.0, you can self-host, fine-tune, and ship commercially. If it is a research license, you cannot use it in production — check before you build.
• Self-hosting shifts the cost and compliance burden to you, but removes per-token fees and data-residency concerns — attractive for EU deployments.

When to choose it

Choose Mistral when European data residency, self-hosting, or open weights matter. For pure managed convenience with the widest ecosystem, OpenAI or Anthropic remain the defaults.