Priority Tier

Priority Tier gives eligible API requests priority queueing for workloads that need more predictable access to shared infrastructure. This page is for enterprise administrators who plan API capacity and developers who add Priority Tier routing to completion requests.

Priority Tier routes eligible API requests through a priority queue before Standard Tier traffic during periods of high load. It is intended for real-time and business-critical workloads that need predictable access to hosted Mistral models.

Priority Tier availability depends on all of the following:

• Your organization has an active Priority Tier entitlement.
• The request uses a model configured for Priority Tier.
• The request is within your custom Priority Tier rate limits for that model.
• Priority Tier capacity is available for the model and deployment region.

If a request is not eligible for Priority Tier, it can still run as Standard Tier traffic depending on the service_tier value you send.

Set the service_tier request parameter on supported completion requests.

If you omit service_tier, the request defaults to standard_only.

import os
from mistralai import Mistral

client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

response = client.chat.complete(
    model="mistral-medium-latest",
    service_tier="auto",
    messages=[
        {
            "role": "user",
            "content": "What is the capital of France?",
        }
    ],
)

print(response.choices[0].message.content)
print(response.usage.service_tier)

import os
from mistralai import Mistral

client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

response = client.chat.complete(
    model="mistral-medium-latest",
    service_tier="auto",
    messages=[
        {
            "role": "user",
            "content": "What is the capital of France?",
        }
    ],
)

print(response.choices[0].message.content)
print(response.usage.service_tier)

The response usage object includes the service tier that processed the request. Use this field to confirm Priority Tier usage, detect fallback to Standard Tier, and monitor unexpected downgrades.

Possible response values are:

Example response excerpt:

{
  "model": "mistral-medium-latest",
  "usage": {
    "prompt_tokens": 22,
    "completion_tokens": 70,
    "total_tokens": 92,
    "service_tier": "priority"
  }
}

{
  "model": "mistral-medium-latest",
  "usage": {
    "prompt_tokens": 22,
    "completion_tokens": 70,
    "total_tokens": 92,
    "service_tier": "priority"
  }
}

Before you send Priority Tier traffic, make sure you have:

• An API key for the organization configured for Priority Tier.
• An active Priority Tier agreement or account setup.
• Custom Priority Tier rate limits configured for each target model.
• A supported SDK version if you use Python or TypeScript.
• Confirmation that your target model is eligible for Priority Tier.

1. Contact your account executive or Mistral contact.
2. Share your target models and expected traffic profile.
3. Wait for Mistral to configure your custom Priority Tier rate limits.
4. Update eligible requests to set service_tier to auto.
5. Check the served tier in the response usage object.

Priority Tier limits are custom and model-specific. Your configured limits can include requests per minute, tokens per minute, or other limits agreed with Mistral.

When a request with service_tier set to auto exceeds your configured Priority Tier limits, the request falls back to Standard Tier instead of failing only because Priority Tier capacity is exhausted.

Priority Tier is billed as a premium multiplier on top of Standard Tier list pricing. The multiplier is 1.75x standard list pricing, which is a 75% premium on input, output, and cached tokens.

Priority requests retain full eligibility for prompt caching discounts, up to 90% savings on cached input tokens. The 1.75x multiplier applies after the prompt cache discount is applied.

Requests served as Priority Tier traffic are billed according to your Priority Tier terms. Requests that fall back to Standard Tier are handled as Standard Tier traffic.

• Regional Inference
• Batch Processing
• SDKs
• API reference