Quick answer
Choose a Mistral Regional Endpoint when an eligible workload needs EU or US inference processing and its model, tools, and features are available in that region. Choose the global endpoint when a specific inference geography is not required or the workload needs a currently unsupported regional feature.
Do not route by hostname alone. First pass the regional readiness checker, then record the compliance, feature, performance, reliability, and cost decision.
Endpoint comparison
| Dimension | Global endpoint | EU or US Regional Endpoint |
|---|---|---|
| Base URL | api.mistral.ai | api.eu.mistral.ai or api.us.mistral.ai |
| Inference location | No specific geography commitment | Selected EU or US geography for eligible inference |
| Price multiplier | Standard list price | 1.1× current list price |
| Model availability | Broader global catalog | Only models hosted in the selected region |
| Agents, Batch, Files | Available according to global product support | Not currently available |
| Regional tools | Global product support | Function calling only in current docs |
| Control plane | Mistral control plane | Still not fully regionalized |
Decision sequence
- Classify inference inputs, outputs, metadata, logs, files, and tool data.
- Decide whether a specific processing geography is mandatory, preferred, or unnecessary.
- Query model availability against the exact regional base URL.
- Reject or redesign workflows using unsupported regional features.
- Compare quality, latency, errors, and cost with representative traffic.
- Log the selected endpoint and verify it per request.
- Review zero-data-retention and downstream systems separately.
Regional proximity can reduce latency for concentrated users, but it is not a latency guarantee. Benchmark full distributions, not one request.
Add Priority Tier deliberately
Priority Tier can be available globally or regionally depending on account setup, model, and capacity. It has its own entitlement, model-specific limits, 99.5% uptime SLA, 1.75× price multiplier, and Standard fallback behavior.
Regional and Priority controls solve different problems. Read the Priority Tier guide and confirm whether price multipliers and contractual terms combine for the exact account.
Frequently asked questions
When should I use a Mistral Regional Endpoint?
Use one when an eligible workload needs a selected EU or US inference-processing geography or users may benefit from regional proximity, and the required model and features are supported there.
What happens when no Mistral region is specified?
The client uses the global endpoint. Mistral does not commit that global-endpoint inference will run in a specific geography.
Are regional endpoints more expensive?
Mistral currently documents regional inference at 1.1× Standard list pricing for input, output, cache reads, and cache writes.
Official sources
Source check: August 12, 2026. Recheck endpoints, regions, prices, model and feature availability, SLA, limits, and account terms.