We offer Fast mode for API customers who want access to faster, more consistent performance on certain models. Below are answers to common questions about how it works, pricing, model availability, rate limits, reliability, policies, and eligibility.

Note: Priority processing was renamed Fast mode on July 30, 2026. You can use either service_tier: priority or service_tier: fast in your API requests. 

Learn more about Fast mode.

Is Fast mode available in all regions?

Availability of Fast mode depends on applicable laws and regulations in each jurisdiction. Please contact your OpenAI representative if you have questions about availability in your region.

How it works

Customers can direct traffic to Fast mode on a per request basis using the existing service_tier parameter, with the option service_tier = "fast".

Tokens served by Fast mode will be billed on a per-token basis, priced at a premium relative to Standard processing rates.

In addition to being configured at the request level, you can also default a project to Fast mode in Project settings > Project Default Service Tier: Fast. You can still override per request. Selecting Fast in your project settings is equivalent to selecting Priority.

How does this interact with Scale Tier?

Scale Tier will remain separate from Fast mode. Requests sent to Fast mode will be billed separately and will not count against your purchased Scale Tier TPM bundles.

Can I automatically send my Scale Tier spillover traffic to Fast mode?

No. Traffic sent to Scale Tier will not automatically spill over to Fast mode.

How is Fast mode billed?

Tokens served by Fast mode will be billed on a per-token basis, priced at a premium relative to Standard processing rates.

Is my annual commitment tied to a specific processing mode?

No. All processing modes count against your annual Enterprise spend commitment.

Do I still get a discount on cached input tokens?

Yes. Eligible cached input tokens receive the same discounts as Standard processing.

How do I view my Fast mode usage and spend?

To view tokens processed by Fast mode (formerly known as Priority processing), go to the Usage dashboard, select Chat Completions or Responses, and select Group by Service Tier.

To view Fast mode cost, go to the Usage dashboard, and select Group by Line Item.

On the Usage dashboard, Fast mode requests for GPT-5.6 and earlier models appear as priority, even when the request specifies fast.

Models

Does Fast mode support long context, fine-tuned models, or embeddings?

GPT-5.6 models support long context in Fast mode. Fast mode does not support fine-tuned models or embeddings.

How do other modalities work with Fast mode?

Fast mode supports the same multimodal capabilities available on Standard. In particular, images can be used as inputs to Fast mode and are processed with the same fast latency.

Will future models be supported?

We plan to offer Fast mode on new GPT models, but we don’t guarantee that every model will be supported.

Rate limits

What are the rate limits?

Fast mode consumption is treated the same as standard API traffic for rate limits.

What are the ramp rate limits?

Fast mode has ramp rate limits to ensure consistently high performance for all customers, while still providing flexible, on-demand pricing. If (a) Fast mode performance is degraded AND (b) a customer’s traffic is ramping too quickly, then some requests may be downgraded to Standard processing instead in rare instances.

The current Fast mode ramp rate limit is defined in our Fast mode documentation.

Best practices for staying within your ramp rate limit

Gradually increase traffic when changing models. For example, if your application is transitioning from a previous snapshot to a new one, use a feature flag to transition traffic over the course of a few hours rather than all at once.

Avoid running large data processing or asynchronous jobs on Fast mode. These jobs can ramp traffic very quickly, and often do not need the improved performance of Fast mode.

If you routinely encounter ramp rate limits, consider purchasing Scale Tier quota instead.

Are ramp rate limits shared across my projects or organizations?

Yes, all of your traffic contributes to the same ramp rate limit.

Policies

What happens if Fast mode is not meeting the latency target?

Fast mode for GPT-6 Astra has no latency SLA. For GPT-5.6 and earlier models, Fast mode and Scale Tier have the same SLA treatment. Eligible Enterprise agreements may provide service credits when latency targets are missed. Contact your account director with questions.

Is Fast mode compatible with data residency?

Yes, subject to model-specific availability. EU data residency is supported for GPT-6.1 Sol, GPT-6 Sol, and GPT-6 Luna, but not GPT-6 Astra.

Is Fast mode compatible with ZDR and the BAA?

Yes, subject to applicable model, endpoint, tool, eligibility, and contractual requirements.

Ultrafast for GPT-6 Astra and GPT-6.1 Sol

Ultrafast is a separate service tier for workloads that need lower-latency GPT-6 Astra and GPT-6.1 Sol responses, such as interactive applications and coding workflows.

Fast mode pricing, rate-limit, and policy guidance applies to requests using service_tier="fast". Ultrafast requests use the separate service_tier="ultrafast" tier.

How do I send an Ultrafast request?

For an organization with Ultrafast access, use the Responses API with:

• Model: gpt-6-astra or gpt-6.1-sol

• Service tier: ultrafast

• Optional request header: OpenAI-Service-Tier: ultrafast

The request header is optional. If you include it, its value must match the service tier in the request body.

For applications that execute tool calls, WebSocket mode lets you reuse a connection across turns and reduce connection overhead.

Can I use the same request header for other service tiers?

When supplied, the OpenAI-Service-Tier header accepts fast or ultrafast. If you change to another service tier, such as default, fast, or flex, update or omit the header so it matches the request.

Can I use Ultrafast in Work or Codex?

Ultrafast is also available in Work and Codex for eligible plans and workspaces. ChatGPT plan eligibility and usage differ from API access.

See: ChatGPT Work and Codex for availability and usage details.