Does Mistral charge extra for regional inference?

Yes. Mistral prices the EU and US regional endpoints at 1.1x the global rate, while the default global endpoint at api.mistral.ai carries no upcharge. Mistral's announcement does not publish a separate price for the Priority Tier, which is still in public preview and sold as a committed service level with custom rate limits.

Which Mistral features do not work on a regional endpoint?

Mistral's docs limit regional endpoints to standard chat completion requests plus function calling, the only tool type supported. Stateful features — Agents, Batch and the Files API — are not available regionally, model availability differs between the EU and US, and control-plane data such as billing and access management is not pinned to the region.

Can you run models that Mistral did not build on Mistral's platform?

Yes, this release starts that. Mistral now serves third-party open models on its own infrastructure, beginning with Z.ai's GLM-5.2, and says more will follow. Those outside models get the same regional controls and service commitments as Mistral's own models, so a team can switch model without switching provider or processing region.

What are European Compute Units?

European Compute Units are Mistral's framework for turning multi-year enterprise commitments into access to infrastructure Mistral builds, targeting up to 1 GW of capacity by 2030. ASML, CMA CGM, Amadeus and Caisse des Dépôts are the named anchor participants. ECUs are a purchasing structure, not something a developer calls from the API today.

Mistral AI · 2026-08-11 · major

Mistral Regional Endpoints — pin inference to Europe or the US

Mistral Regional Endpoints are now generally available, letting API customers keep inference inside the EU or the US. The same announcement opens Mistral's platform to outside open models, starting with Z.ai's GLM-5.2.

Mistral AI cover graphic for its sovereign AI infrastructure announcement

Mistral API calls can now be pinned to European or US data centres, and outside open models run on the same platform.

Quick facts

Regional Endpoints	Generally available
Regions	Global, EU (api.eu.mistral.ai), US (api.us.mistral.ai)
Regional pricing	1.1x the global rate
Priority Tier	Public preview, custom rate limits + uptime SLA
Third-party models	Z.ai GLM-5.2, with more planned
Not regional yet	Agents, Batch and Files APIs
Compute coalition	European Compute Units, up to 1 GW by 2030

What is it?

Regional Endpoints move from preview to general availability, so a Mistral customer can decide whether a request is processed in Europe or the United States instead of hitting the global endpoint. Alongside it Mistral opened its serving platform to third-party open models — Z.ai's GLM-5.2 is the first — and put a Priority Tier into public preview for teams that need a committed service level.

How does it work?

Each region group maps to its own base URL, `api.eu.mistral.ai` or `api.us.mistral.ai`, and the Python SDK from version 2.70 selects one with a `server` parameter. Requests routed that way stay inside the chosen geography. The coverage is deliberately narrow for now: standard chat completions and function calling work, while stateful features such as Agents, Batch and the Files API do not, and control-plane data like billing and access management is not region-pinned.

Why does it matter?

European teams that could not send regulated data to a global endpoint get a supported path without leaving the Mistral API, at a stated 1.1x premium over global pricing. Hosting GLM-5.2 also matters for a different reason: a strong Chinese open-weight coding model becomes callable under the same regional controls and service commitments as Mistral's own, so choosing a model no longer forces a change of provider or jurisdiction.

Who is it for?

European enterprises and regulated teams building on the Mistral API

Frequently asked questions

Does Mistral charge extra for regional inference?: Yes. Mistral prices the EU and US regional endpoints at 1.1x the global rate, while the default global endpoint at api.mistral.ai carries no upcharge. Mistral's announcement does not publish a separate price for the Priority Tier, which is still in public preview and sold as a committed service level with custom rate limits.
Which Mistral features do not work on a regional endpoint?: Mistral's docs limit regional endpoints to standard chat completion requests plus function calling, the only tool type supported. Stateful features — Agents, Batch and the Files API — are not available regionally, model availability differs between the EU and US, and control-plane data such as billing and access management is not pinned to the region.
Can you run models that Mistral did not build on Mistral's platform?: Yes, this release starts that. Mistral now serves third-party open models on its own infrastructure, beginning with Z.ai's GLM-5.2, and says more will follow. Those outside models get the same regional controls and service commitments as Mistral's own models, so a team can switch model without switching provider or processing region.
What are European Compute Units?: European Compute Units are Mistral's framework for turning multi-year enterprise commitments into access to infrastructure Mistral builds, targeting up to 1 GW of capacity by 2030. ASML, CMA CGM, Amadeus and Caisse des Dépôts are the named anchor participants. ECUs are a purchasing structure, not something a developer calls from the API today.

Try it

client = Mistral(api_key=os.environ["MISTRAL_API_KEY"], server="eu")