Real-time inference on DeepInfra is fast — but when a popular model is under heavy load, requests queue up and some get shed with an HTTP 429. The new Priority service tier lets your latency-critical traffic jump to the front of that queue and stay admitted through contention, for 1.5× the real-time price. It's a single OpenAI-compatible field on the request — no separate endpoint, no new API to learn.

Why Priority?

Most traffic is happy to retry a 429 and move on. Some isn't. Priority is built for the workloads where waiting in line is the problem:

• Interactive, user-facing apps where time-to-first-token is the product — chat, autocomplete, voice.
• Agentic and multi-step pipelines where every hop's latency compounds into a slow end-to-end run.
• Revenue-critical traffic you can't afford to have shed during peaks.

You opt in per request by setting service_tier to "priority". Leave it off and your request runs at the standard real-time rate, exactly as it does today — nothing changes for traffic that doesn't need to skip the line.

How It Works

One field. Add "service_tier": "priority" to any chat or completions request. Priority requests:

• Jump to the front of the engine's scheduling queue — lower time-to-first-token when the model is busy.
• Get protected admission — they keep being accepted while normal-tier traffic begins to see 429s under contention.
• Echo the tier back — the response's service_tier field comes back "priority" when (and only when) priority was actually applied.

When the model is idle, priority and normal requests look the same — there's no queue to jump. The difference shows up exactly when it matters: under load.

Supported Endpoints

• /v1/chat/completions
• /v1/completions

Priority-rated billing also applies to embeddings. Everything works through the standard OpenAI-compatible API you're already using.

Pricing

Priority is billed at 1.5× the corresponding real-time price — applied automatically, with no extra configuration.

Here's the part worth reading twice: you only pay the priority rate when priority is actually delivered. If you request priority on a model that doesn't support it, the request is served normally, billed at the normal rate, and the response's service_tier comes back "default". What's billed always equals what's echoed — so you can verify exactly what you paid for by reading the service_tier field on the response. No silent upcharge for a tier you didn't get.

Supportability Today

Priority is live now on models served on our vLLM stack, which covers the bulk of our text-generation catalog. Support for models served on SGLang and TensorRT-LLM is rolling out.

Priority is enabled on a per-model basis, and the set of priority-enabled models is growing continuously. You don't have to guess which ones qualify: every model that supports priority already carries a Priority tag on its model page, so you can see at a glance whether a model honors the tier before you send a request.

And you can always confirm it programmatically: send a request with service_tier="priority" and check the echoed service_tier on the response. If it comes back "priority", you're at the front of the queue; if it comes back "default", the model isn't priority-enabled yet and you weren't charged the premium.

Get Started

Using the OpenAI Python client — just add service_tier="priority" and read it back off the response:

from openai import OpenAI

client = OpenAI(
    api_key="$DEEPINFRA_TOKEN",
    base_url="https://api.deepinfra.com/v1/openai",
)

resp = client.chat.completions.create(
    model="meta-llama/Meta-Llama-3.1-8B-Instruct",
    messages=[{"role": "user", "content": "Summarize this support ticket: ..."}],
    service_tier="priority",
)

print(resp.choices[0].message.content)
print("served as:", resp.service_tier)  # "priority" when priority was applied
copy

Or with curl — service_tier is just another field in the JSON body:

curl https://api.deepinfra.com/v1/openai/chat/completions \
  -H "Authorization: Bearer $DEEPINFRA_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
    "messages": [{"role": "user", "content": "Summarize this support ticket: ..."}],
    "service_tier": "priority"
  }'
copy

The service_tier field on the response tells you which tier actually served the request.

Skip the line

See the Service Tier documentation for the full reference, and start sending your latency-critical traffic to the front of the queue.