Mistral is taking three steps toward what it calls AI sovereignty. The French AI company is making Regional Endpoints generally available, introducing a Priority Tier with an SLA, and beginning to support open models from third parties for the first time. This starts with GLM-5.2 from the Chinese company Z.ai.
Until now, Mistral only ran Mistral models. That’s about to change: customers will soon be able to use open models from other providers as well, starting with GLM-5.2 from Z.ai. According to Mistral, these models run on the same infrastructure, with the same regional controls and service commitments as its own models.
The reasoning behind this is that companies no longer build their AI stack around a single model. Mistral refers to a collection of capabilities: advanced reasoning, high-volume production, and work tailored to an organization’s own data. In addition to state-of-the-art models, the company offers specialized variants such as Mistral OCR and Voxtral, as well as models trained on corporate data.
Region selection and an SLA
Mistral Regional Endpoints are now generally available. This lets customers choose whether their inference runs in Europe or the United States, aligning with requirements for data residency, regulations, and latency. Processing takes place in the selected region, with limited and restricted exceptions for sub-processors outside that region.
In addition, Mistral is introducing a Priority Tier, currently in public preview. It offers guaranteed service levels for mission-critical workloads, including customized rate limits and an uptime SLA. Mistral claims to be the only European AI lab that combines both.
Tip: Microsoft and Mistral expand AI collaboration in Europe