Mistral has launched regional inference, allowing customers to process AI requests within the EU for a 10% premium. However, this feature comes with significant limitations, as only function calling is supported at regional endpoints.
TL;DR
- Mistral introduces regional inference with EU data processing for a 10% fee.
- Priority Tier offers faster queue access with a 75% surcharge and a 99.5% uptime SLA.
- Mistral opens its platform to third-party models, starting with GLM-5.2 from Z.ai.
What happened
Mistral has made regional inference generally available, enabling customers to route AI requests through either European (api.eu.mistral.ai) or US endpoints. This regional processing ensures data stays within the chosen region, crucial for compliance and lower latency, but costs 10% more than standard pricing.
The EU data processing feature has notable limits. Only function calling is available at regional endpoints, while agents, batch processing, and file management are not supported. Model selection also varies by region, requiring customers to query each endpoint to see available models. Additionally, account settings and API keys may still be processed outside the chosen region, and limited transfers to subcontractors are allowed.
Mistral has also launched the Priority Tier, currently in open beta, which offers faster queue access during peak times. This tier includes a 99.5% uptime SLA and is targeted at use cases where latency is costly. The Priority Tier costs 1.75x the standard price, with discounts from prompt caching applied before the surcharge. Customers must sign a contract with Mistral's sales team to activate priority access.
Why it matters
For developers and startups, the regional inference feature provides a way to comply with data residency requirements, but the lack of support for stateful features like agents and file management may limit its usefulness for some applications. The Priority Tier offers a way to ensure faster response times, which can be critical for production systems.
Investors should note that Mistral is expanding its platform by hosting third-party models, starting with GLM-5.2 from Z.ai. This move aligns with Mistral's involvement in the Open Secure AI Alliance and Nvidia's Nemotron coalition, potentially broadening its market reach and revenue streams. However, the success of this strategy depends on securing long-term commitments from large customers.
Key facts
- Regional inference is available at api.eu.mistral.ai and api.us.mistral.ai.
- EU data processing costs 10% more than standard pricing.
- Only function calling is supported at regional endpoints.
- Priority Tier offers a 99.5% uptime SLA and costs 1.75x the standard price.
- Priority Tier is not self-service and requires a contract with Mistral's sales team.
- Mistral is hosting third-party models, starting with GLM-5.2 from Z.ai.
- Mistral is a member of the Open Secure AI Alliance and Nvidia's Nemotron coalition.
- European Compute Units are multi-year purchase commitments to fund new data centers in Europe.
Context
Mistral's regional inference feature is a response to growing concerns about data sovereignty and latency in AI processing. By offering EU data processing, Mistral caters to industries with strict data residency requirements, such as banking and government agencies. However, the limitations on stateful features highlight the challenges of balancing compliance with functionality.
The Priority Tier is part of Mistral's strategy to offer tiered services, similar to other cloud providers. This approach allows Mistral to target different market segments, from cost-conscious startups to enterprises willing to pay for guaranteed performance. The inclusion of third-party models further expands Mistral's platform, potentially attracting a broader range of customers.