Quick answer
Mistral Priority Tier is a public-preview, account-configured service for eligible real-time and business-critical requests. It uses priority queueing, custom limits by model, a documented 99.5% uptime SLA, and a 1.75× list-price multiplier.
It is not an always-priority switch. Eligibility depends on organization entitlement, model configuration, the model’s custom Priority limits, and capacity in the deployment region. Use the regional readiness checker to test the complete plan.
Eligibility gate
A request needs all of the following:
- an active Priority Tier entitlement for the organization;
- a target model configured for Priority Tier;
- traffic within the model-specific custom limits;
- available Priority capacity for the model and deployment region;
- a supported completion request using the intended service-tier value.
Mistral instructs customers to coordinate target models and traffic profiles with their account contact. Do not infer entitlement from documentation access or an API key.
auto versus standard_only
Set service_tier to auto only when a request should use Priority if eligible and may fall back to Standard. Use standard_only, or leave the parameter unset, for Standard traffic.
The response usage object reports whether the request was served as priority or standard. Log that field with the model, endpoint, region, latency, limit state, and request identifier. A successful response does not prove it consumed Priority capacity.
SLA, limits, and cost
| Dimension | Priority Tier boundary |
|---|---|
| Uptime SLA | 99.5% in current docs |
| Limits | Custom and model-specific |
| Limit behavior | auto can fall back to Standard |
| Price | 1.75× Standard list price |
| Prompt caching | Discount can still apply; multiplier follows the discounted input price |
| Availability | Depends on entitlement, model, account setup, region, and capacity |
An uptime SLA does not guarantee per-request latency, no fallback, unlimited burst, or feature availability. Define separate SLOs for served-tier rate, latency percentiles, errors, fallback, quality, and cost.
For endpoint selection, compare regional and global inference. For data-location boundaries, read the EU/US regional endpoint guide.
Frequently asked questions
What does Mistral Priority Tier provide?
It gives eligible real-time requests priority queueing on shared infrastructure, custom model-specific rate limits, and a documented 99.5% uptime SLA.
How much does Priority Tier cost?
Mistral currently documents a 1.75× multiplier on Standard list pricing for input, output, and cached tokens. Standard fallback traffic is handled as Standard Tier.
Can a Priority request fall back to Standard Tier?
Yes. With service_tier set to auto, an ineligible or capacity-limited request can run as Standard Tier. Log the served tier in the response usage object.
Official sources
Source check: August 12, 2026. Recheck public-preview status, SLA terms, entitlement, models, limits, regions, caching, and price before use.