Mistral has introduced options for customers to route AI requests through servers in Europe or the US, along with priority access during peak traffic. Both services come with additional costs and constraints. The decision reflects Mistral's broader European AI sovereignty strategy, allowing companies to process data within specific regions. Customers can choose between European and US endpoints, with regional processing important for industries requiring data localization, such as banks and insurers. Processing within a region also reduces latency due to shorter network paths. However, the default endpoint does not guarantee regional processing, and customers must opt for the regional endpoint to benefit from these advantages.
Regional routing incurs a 10 percent surcharge on top of standard pricing. While EU data processing is available, it has notable limitations. Only function calling works with regional endpoints, restricting the model's ability to trigger external APIs. Features like agents, batch processing, and file management are not available through regional addresses. Model selection also varies by region, with customers needing to query each endpoint to see available models. The company attributes these limitations to the need for persistent storage, which is not required for standard model calls. The scope of regional processing is narrower than the term 'sovereignty' suggests, as account settings, API keys, and billing can still be processed outside the chosen region.
The blog post also mentions limited, secured transfers to subcontractors outside the region. What is regional is the compute step, not the entire platform. Whether requests are stored or logged afterward depends on a separate setting called Zero Data Retention. This means customers sending contract text to a model can use the EU endpoint, but those needing agents or file management through the Files API won't receive the same guarantees. Mistral is also offering a Priority Tier, currently in open beta, to address latency issues during high traffic. The Priority Tier provides faster processing for paying customers, with an uptime SLA of 99.5 percent, allowing about three and a half hours of downtime per month. Standard customers have no such guarantee. Priority access is activated through an API parameter called service_tier, with the default set to 'standard_only'.
Source: thedecoder
