Mistral

Mistral is giving customers the option to route AI requests through servers in either Europe or the US, and selling priority queue access during peak traffic. Both come with a surcharge, and the regional routing doesn't cover all features or data.

When companies build Mistral's AI into their own products, requests go to one of the provider's servers. Two questions matter for enterprise customers: where that server actually sits, and what happens when everyone sends requests at once. Mistral is now selling answers to both, framing them in a blog post as part of a broader European AI sovereignty strategy.

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), and processing stays in that region. That matters for banks, government agencies, and insurers that need to prove customer data never leaves the EU. Shorter network paths also mean lower latency. Anyone using the default endpoint gets no guarantee about where their request is processed. Regional routing costs 10 percent on top of standard pricing.

EU data processing comes with significant limits

However, among the platform's add-on tools, only function calling works with regional endpoints, meaning the model's ability to trigger external APIs. Agents, batch processing, and file management aren't available at the regional addresses. Model selection varies by region, too. Mistral doesn't publish a fixed list, customers have to query each endpoint to see what's there.

The likely reason is the gap between a simple model query and storing intermediate state. A standard model call needs no persistent storage. Agents, batch jobs, and file storage hold data beyond a single call, things like intermediate steps or uploaded documents. Mistral calls these features "stateful." They probably require extra on-site infrastructure, though the company hasn't confirmed that.

The scope is also narrower than the word "sovereignty" suggests. Account settings, API keys, billing, and usage stats can still be processed outside the chosen region, according to the documentation. The blog post also mentions limited, secured transfers to subcontractors outside the region. What's regional is the compute step, not the whole platform. Whether requests get stored or logged afterward depends on a separate setting called Zero Data Retention.

What this means in practice: customers who send contract text directly to a model can run it through the EU endpoint. Anyone who needs agents or file management through the Files API won't get the same guarantee.

Why companies would pay for a faster queue

The second offering is the Priority Tier, currently in open beta. All customers share the same data centers. When lots of requests hit at once, response times go up. The Priority Tier is a fast lane: paying customers' requests get processed ahead of regular traffic when things get busy. Mistral is targeting use cases where latency costs real money, like a customer service chatbot or a production system on a factory floor.

The tier includes an uptime SLA of 99.5 percent, a contractually guaranteed service level that allows roughly three and a half hours of downtime per month. Mistral's standard tier has no such guarantee.

Customers activate priority access through a single API parameter called service_tier. Setting it to "auto" sends the request through the fast lane when capacity is available. The default value is "standard_only," which takes the regular path. Each customer also gets individually negotiated rate limits for how many requests per minute get priority treatment. Going over that limit doesn't fail the request. It just falls back to standard processing. The API response shows which tier actually handled the request, so customers can check whether they're getting what they pay for.

Mistral charges 1.75x the standard price, a 75 percent surcharge. Discounts from prompt caching, where repeated text segments are stored and billed at lower rates, still apply. Those discounts can reach 90 percent and get calculated first according to the documentation, with the priority surcharge applied after. The Priority Tier isn't self-service. Customers have to sign a contract with Mistral's sales team.

Third-party models join the platform

Mistral is also opening its platform to open models from other providers. First up is GLM-5.2 from Chinese AI company Z.ai, which runs under the same regional rules and guarantees as Mistral's own models. To fund the compute capacity this requires, Mistral is collecting multi-year purchase commitments from large customers, packaged as European Compute Units. The idea is that building new data centers in Europe only pencils out if enough companies commit long-term.

Mistral is a member of the Open Secure AI Alliance and Nvidia's Nemotron coalition. The company sees hosting third-party model weights on its platform as a natural extension of that work.

Read on for the full picture.
Subscribe for hype-free coverage.

• Full access to every article on THE DECODER
• No ads
• Join the comments and community discussions
• A weekly AI news recap via mail
• 6x/year: "AI Radar" — deep dives on the AI topics that matter most
• Daily AI news, always up to date
• Our full ten-year archive
• Covered by a team with 10+ years in AI