service mistralvendor Mistral AIcategory Model APIs
Depends (Caution)Free mode: may train. Pay-as-you-go: you can opt out at any time. Vibe (chat): trains by default unless you opt out. Enterprise: opted out by default.
Free mode: may train. Pay-as-you-go: you can opt out at any time. Vibe (chat): trains by default unless you opt out. Enterprise: opted out by default.
As stated during subscription, we may use your data (input and output) to train our artificial intelligence models.
Standard API inputs and outputs are kept for 30 rolling days for abuse monitoring unless ZDR is on. The Agents API keeps data until account termination, and fine-tuning data until you delete it.
Except for specific APIs, we keep your Input and Output for the period necessary to generate the Output and then for thirty (30) rolling days to monitor abuse (unless zero data retention is activated).
Against the 21 tools that state a period (grey dots):
Pay-as-you-go only, stateless endpoints only, approved at Mistral's discretion. Files, batch, agents and conversations are out of scope.
We will review your request and, at our discretion, approve or deny it.
The DPA lists automated moderation and API abuse monitoring as processing purposes, except when ZDR is active.
Automated moderation, including abuse monitoring on our APIs (except, in this last case, when zero data retention has been activated), to enforce the Agreement.
EU by default; US via a separate endpoint. Some features may transfer data outside the EU to listed subprocessors.
By default, your data is hosted in the European Union.
Reports available on request via the Trust Center.
Yes, Mistral complies with both SOC 2 Type II and ISO 27001/27701 frameworks.
Not stated on Mistral's public pages read (help center, DPA, trust center front page, which lists ISO 27001, ISO 27701 and SOC 2 Type II only). Third-party sites say a BAA exists for some services, but this could not be confirmed on an official page.
no official page stated this clearly
Mistral documents self-deploying its models with vLLM, TensorRT-LLM or TGI.
Mistral AI models can be self-deployed on your own infrastructure through various inference engines.
Spotted something wrong or out of date? Corrections are welcome; see how to send one. How statuses are assigned.