Mistral is giving enterprise AI customers more control over two practical questions: where model requests are processed, and how quickly they move through the system when traffic is high. The new options are useful, but they are not blanket guarantees across the full platform.
The company is positioning regional inference, Priority Tier access, and third-party model hosting as part of a broader European AI sovereignty strategy. For customers in sectors such as banking, government, and insurance, the details matter because the limits are as important as the headline features.
Regional inference adds Europe and US endpoints
Regional inference is now generally available. Customers can send requests to a European endpoint, api.eu.mistral.ai, or a US endpoint, api.us.mistral.ai. Mistral says processing stays in the selected region when those endpoints are used.
That gives customers a clearer answer to a basic infrastructure question: where is the server that handles the request? For organizations that need to show customer data never leaves the EU, the European endpoint can be an important option. Shorter network paths can also reduce latency.
The default endpoint does not provide the same regional guarantee. Customers that want regional routing pay 10 percent on top of standard pricing.
In practical terms, this makes regional inference a paid control rather than the normal behavior of the platform. It can help when a company sends text directly to a model and wants that processing step to happen in Europe or the US, but the protection does not automatically extend to every feature around the model.
The EU data processing promise has boundaries
The main limitation is feature coverage. Among Mistral’s platform add-on tools, only function calling works with regional endpoints. Function calling is the model capability that lets it trigger external APIs.
Other tools are not available at the regional addresses. Agents, batch processing, and file management are outside the current regional endpoint support. Model availability can also differ by region, and Mistral does not publish a fixed model list for each endpoint. Customers have to query each endpoint to see what is available.
The reason appears to involve the difference between a single model request and features that keep state over time. A basic model call does not require persistent storage. Agents, batch jobs, and file storage can hold information beyond one request, including intermediate steps or uploaded documents. Mistral calls these features "stateful." The company has not confirmed whether extra on-site infrastructure is the reason for the gap.
There is also a narrower definition of sovereignty than some buyers may expect. Account settings, API keys, billing, and usage stats can still be processed outside the chosen region, according to the documentation. Mistral’s blog post also refers to limited, secured transfers to subcontractors outside the region.
That means the regional guarantee is focused on compute, not the entire customer relationship with the platform. Whether requests are stored or logged after processing depends on a separate setting called Zero Data Retention.
For a simple workflow, such as sending contract text directly to a model, the EU endpoint can provide regional processing for that request. For workflows that need agents or file management through the Files API, the same regional guarantee is not available.
Priority Tier sells a faster path during peak traffic
Mistral’s second new offering is Priority Tier, which is currently in open beta. It is designed for the moments when many customers send requests at the same time and response times rise.
All customers still use the same data centers. Priority Tier changes the order of processing under load: paying customers’ requests are handled ahead of regular traffic when the system is busy.
Mistral is aiming this at use cases where latency has direct operational impact. The source examples are a customer service chatbot and a production system on a factory floor. In both cases, slower responses can affect the quality or reliability of the product that depends on the AI system.
The tier also includes an uptime SLA of 99.5 percent. That service level allows roughly three and a half hours of downtime per month. Mistral’s standard tier does not include that guarantee.
How priority access is activated and billed
Customers enable priority handling with a single API parameter: service_tier. Setting service_tier to "auto" sends a request through the faster lane when priority capacity is available. The default value is "standard_only," which uses regular processing.
Priority access is not unlimited. Each customer receives individually negotiated rate limits for how many requests per minute can receive priority treatment. If a customer exceeds that limit, the request does not fail. It falls back to standard processing instead.
Mistral also returns information in the API response showing which tier actually processed the request. That gives customers a way to verify whether traffic received priority handling.
The price is higher. Mistral charges 1.75x the standard price, which is a 75 percent surcharge. Prompt caching discounts still apply, where repeated text segments are stored and billed at lower rates. Those discounts can reach 90 percent and are calculated first, with the priority surcharge applied after.
Priority Tier is not self-service. Customers must sign a contract with Mistral’s sales team.
Third-party models expand the platform
Mistral is also opening its platform to open models from other providers. The first named model is GLM-5.2 from Chinese AI company Z.ai. It runs under the same regional rules and guarantees as Mistral’s own models.
To support the compute capacity needed for this expansion, Mistral is seeking multi-year purchase commitments from large customers. These are packaged as European Compute Units. The logic is straightforward: new data centers in Europe require enough long-term customer commitment to justify the investment.
Mistral is a member of the Open Secure AI Alliance and Nvidia’s Nemotron coalition. The company views hosting third-party model weights on its platform as an extension of that work.
The overall shift gives customers more choice, but also more to evaluate. Regional endpoints, Zero Data Retention, stateful feature support, rate limits, and negotiated contracts all shape what Mistral’s new options mean in production. For enterprise buyers, the key question is not only whether a feature exists, but exactly which parts of a workflow it covers.