Does APIYI Support OpenAI Fast / Ultrafast Mode?

APIYI does not currently support OpenAI service_tier levels (Fast, Ultrafast, Priority, Flex). Passing one does not error, but the request is processed and billed at the standard tier.

Short Answer

Not at the moment. A request that includes service_tier (fast, ultrafast, priority, flex) does not error and still returns 200, but it is actually processed at the standard tier (default):

• Speed is the same as without the parameter: no faster, no slower
• Billing uses the model’s standard price, with no surcharge and no discount
• /v1/responses returns service_tier as "default"; /v1/chat/completions does not include the field

So there is no need to send this parameter for now, and sending it never costs extra.

What These Tiers Are

OpenAI uses the service_tier parameter to select a processing tier:

Tier	Request value	OpenAI’s description
Fast (formerly Priority)	fast; the old value priority still works	Renamed from Priority on 2026-07-30; faster, priced above the standard tier
Ultrafast	ultrafast	The fastest tier; only gpt-6.1-sol and gpt-6-astra, Responses API only, priced well above the standard tier
Flex	flex	Slower and cheaper; returns 429 when provider-side capacity is short
Standard	Omitted, or auto / default	The default tier

None of these tiers currently takes effect on APIYI. Every request is handled at the standard tier.

Test Results

On 2026-10-13 we tested gpt-6.1-sol and gpt-6-luna in both the default group and the OpenAI official group. Both endpoints and all four tiers returned 200:

Observation	Result
service_tier returned by /v1/responses	Always default
service_tier returned by /v1/chat/completions	Field not returned
Latency and output speed	No difference from requests without the parameter
Console billing	Standard price, identical to requests without the parameter

If your code already sends service_tier: "priority" or "auto", you can leave it as is. It affects neither the call nor the billing.

If You Need It

If your workload genuinely needs the low latency of Fast or Ultrafast, contact support with your use case and expected volume. We will evaluate whether to enable it and announce the billing method if we do.

Related Docs

What Does an OpenAI Model Refusal Look Like?

GPT refusals return 200 with no error code or category

How Do I Avoid API Timeouts?

Reasoning models take longer, so leave enough client timeout