Mistral announced the offering of regional data processing in Europe and the United States, priority access at peak times, and the inclusion of third-party open models on its platform. The first two features have surcharges and limitations, the company said.
In regional processing, requests can be sent to the European endpoint (api.eu.mistral.ai) or the US endpoint (api.us.mistral.ai), with the guarantee that computing occurs in that region. The service is aimed at banks, public agencies, and insurers that need to prove customer data does not leave the EU, and it also reduces latency.
The option costs 10% more than the standard price. Those who use the standard endpoint have no guarantee about where the request is processed.
Regional processing limits
Among the platform's additional tools, only function calling, which allows the model to invoke external APIs, works with the regional endpoints. Agents, batch processing, and file management are not available at these addresses. Model selection also varies by region, with no fixed published list.
Mistral did not confirm the reason, but the difference probably lies in state storage: agents, batches, and files store data beyond a single call. Moreover, according to the documentation, account settings, API keys, billing, and usage statistics may still be processed outside the chosen region. The company also mentions limited and secure transfers to subcontractors outside the region.
In practice, customers who send contract text directly to the model can use the European endpoint. Those who depend on agents or the file API will not have the same guarantee.
Priority costs 75% more
Priority access, called Priority Tier, is in open beta. All customers share the same data centers, and the feature creates a fast lane so that paid requests are processed before others during congestion. The target audience is cases where latency causes financial loss, such as customer service chatbots or factory floor systems.
The tier includes an availability SLA of 99.5%, about three and a half hours of downtime per month, something the standard plan does not offer. Activation is done through the service_tier parameter; the value "auto" sends the request through the fast lane when capacity is available, while the default is "standard_only". Per-minute request limits are negotiated individually, and the excess goes back to normal processing without failing.
The surcharge is 75% on top of the standard price, totaling 1.75 times the value. Prompt caching discounts, which can reach 90%, are calculated before the surcharge. Priority Tier is not self-service: it requires a contract with Mistral's sales team.
Third-party models on the platform
Mistral also opened the platform to open models from other providers. The first is GLM-5.2, from Chinese Z.ai, which follows the same regional rules and guarantees as Mistral's models. To finance computing capacity, the company collects multi-year commitments from large customers, packaged as European Compute Units. The idea is to enable new data centers in Europe with long-term commitments.
