Azure Priority Processing

Azure sells a low-latency lane for its OpenAI models, but until now the gateway stripped the field that selects it — a Priority request routed through LLM Gateway quietly ran at standard speed. Azure Priority processing is now a first-class service tier, requested exactly the way you already request it on OpenAI.

The same field you already use

Azure accepts the OpenAI-compatible service_tier body field on its v1 chat completions and responses endpoints, with the same values. There is no Azure-specific parameter to learn: point the model string at an Azure mapping that offers the tier and send service_tier: "priority".

1curl https://api.llmgateway.io/v1/chat/completions \2  -H "Authorization: Bearer $LLM_GATEWAY_API_KEY" \3  -H "Content-Type: application/json" \4  -d '{5    "model": "azure/gpt-6-sol",6    "service_tier": "priority",7    "messages": [{ "role": "user", "content": "Summarize this incident report." }]8  }'

1curl https://api.llmgateway.io/v1/chat/completions \2  -H "Authorization: Bearer $LLM_GATEWAY_API_KEY" \3  -H "Content-Type: application/json" \4  -d '{5    "model": "azure/gpt-6-sol",6    "service_tier": "priority",7    "messages": [{ "role": "user", "content": "Summarize this incident report." }]8  }'

Azure offers Priority only — it publishes no Flex rate card — so a Flex request is still rejected for Azure rather than silently downgraded. Which models carry the tier, and the premium each one bills, are listed on the models page.

Downgrades are billed at what actually ran

Priority is a lane, not reserved capacity. Azure serves a Priority request at standard when your subscription is not entitled to the tier, during peak load, when traffic ramps faster than the rate limit allows, and for long-context prompts on some models.

Azure reports the tier it actually served, and the gateway bills that rather than the tier you requested. A downgraded request is charged at standard rates and comes back with used_service_tier: null, next to the requested_service_tier you sent, so the difference is visible per request in your logs instead of showing up as an unexplained premium on your invoice.

Retries and fallback never downgrade a tier you asked for: routing is narrowed to mappings that can serve it before a provider is picked, so a request that cannot run at Priority fails with a 400 instead of quietly running at standard.

Service tiers docs → | Azure integration guide →