Features
Latency & SchedulingService Tiers
Every request runs on the standard tier unless you ask for another one. On supported models you can request Priority processing for an individual request by setting a single parameter.
Priority processing
Priority requests are scheduled ahead of standard requests for the same model. This shortens waiting time when a model is under heavy demand, which helps interactive and latency-sensitive workloads such as chat interfaces, coding agents, and multi-step pipelines where every call adds to the total time. When a model is not busy, response times on both tiers are similar.
Priority is available on selected models. A model that supports it shows its Priority prices on its model page.
Requesting Priority
Add "service_tier": "priority" to the request body.
from openai import OpenAI
client = OpenAI(
api_key="YOUR_DEVUP_API_KEY",
base_url="https://api.devupai.com/v1",
)
response = client.chat.completions.create(
model="deepseek-ai/DeepSeek-V4-Flash",
service_tier="priority",
messages=[{"role": "user", "content": "Hello!"}],
)
print(response.service_tier)
print(response.choices[0].message.content)If your SDK version does not accept service_tier as a named argument, pass it through the SDK's extra-body option, for example extra_body={"service_tier": "priority"} in Python.
Checking which tier served your request
Every response includes a top-level service_tier field with the tier that actually processed the request: "priority" or "default".
{
"object": "chat.completion",
"model": "deepseek-ai/DeepSeek-V4-Flash",
"service_tier": "priority",
"choices": [ ... ],
"usage": { "prompt_tokens": 12, "completion_tokens": 48, "total_tokens": 60 }
}Streaming. When stream is true, the chunks carry service_tier. The first chunk may report null. Use the value from the last chunk.
Billing
- Priority requests are billed per 1M tokens at the model's Priority prices: input, output, and cached input. They are shown on each model's page.
- You are billed at Priority prices only when the response reports
"service_tier": "priority". If a request is processed on the standard tier, it is billed at standard prices. - As with every request,
_devup.cost_dzdin the response shows the amount charged.
Supported values and fallback
| Value | Behaviour |
|---|---|
| omitted | Standard tier, standard prices. |
| "priority" | Priority on models that support it. On other models, the request is processed on the standard tier at standard prices, with no error. |
| any other value | Processed on the standard tier at standard prices. |
Priority is available on the Chat Completions endpoint.
When to use Priority
- Use Priority for user-facing requests where waiting is the problem: chat UIs, voice, autocomplete, agent steps.
- Keep standard for background work such as evaluations, data enrichment, and batch jobs.
- Read
service_tierin the response to confirm the tier and reconcile your costs.