Notes

Mistral confirms its free plan trains models on what you write to it


The updated documentation clarifies that free-tier conversations feed training unless the account turns it off manually.

September 3, 2026 · Translated from the Spanish original

What happened

Why it matters

The number

2 separate switches. Turning off training in Vibe doesn’t turn it off for the API, or vice versa. You have to do it twice, in different places.

Context

The correction circulated on Hacker News, where several developers pointed out that the new text contradicts the previous wording. Another page of Mistral’s documentation still states that data sent through the API isn’t used for training, without distinguishing the free tier from the paid one. Both statements coexist on the same site today, and neither refers to the other. In the week Brussels required another provider to audit how it works on the inside, documentation was still the only place to look.

What’s next

Bottom line

None of these conditions changed this week. What changed is that they’re now written where someone might look for them, and the data that went in before the correction went in all the same.

This note describes terms of service published by a provider and does not constitute legal advice.

Sources

Related notes

← All notes