OpenAI Ultrafast vs Fast vs Standard API Tiers
On this page
Quick answer
Choose the service tier from measured business need:
| Tier | Current boundary | Verify in response |
|---|---|---|
| Standard | Default pricing and performance for the selected model | service_tier: "default" |
| Fast | Premium processing, formerly Priority; up to 2.5× Standard for GPT-5.6 Sol | service_tier: "priority" for GPT-5.6 and earlier |
| Ultrafast | Access-controlled gpt-5.6-sol limited preview; up to 14× Standard vendor claim | service_tier: "ultrafast" |
Do not compare speed multipliers as if they were guaranteed end-to-end percentiles. Test the same workload and calculate cost per accepted task.
Standard
Standard is the baseline. It is the right control for offline work, flexible latency, and workloads where premium speed does not change user or business outcomes.
Fast
OpenAI renamed Priority processing to Fast mode. Requests may specify fast or priority, while GPT-5.6 and earlier responses report priority when served through Fast.
Fast shares model rate limits with Standard. OpenAI also documents a ramp-rate rule: traffic of at least 1 million tokens per minute that rises by more than 50% within 15 minutes can be served at Standard speed and charged Standard rates. Verify service_tier: "default" in those responses and ramp gradually.
Fast has published pricing and SLA guidance. It is separate from Scale Tier, and it does not support fine-tuned models or embeddings in the current guide.
Ultrafast
Ultrafast is not simply “faster Fast.” It is an access-controlled preview for gpt-5.6-sol. OpenAI’s announcement reports peak output speed and selected use cases but does not establish a public price, universal capacity, or general-access date.
For production design, obtain current access, price, limit, region, capacity, fallback, and SLA terms directly. Require explicit fallback behavior when Ultrafast is unavailable or fails acceptance tests.
Decision metrics
Compare time to first token, output tokens per second, end-to-end p50/p95/p99, tool time, task success, cost per request, cost per accepted task, concurrency, errors, timeouts, fallback rate, and downstream action correctness.
Use the Ultrafast preview guide for source boundaries and the readiness check for a controlled pilot.
Frequently asked questions
What is the difference between Standard, Fast, and Ultrafast?
Standard is default API processing; Fast is premium priority processing with documented billing and ramp behavior; Ultrafast is an access-controlled limited-preview tier currently documented for gpt-5.6-sol.
What response tier does Fast mode return?
For GPT-5.6 and earlier models, OpenAI says requests specifying fast or priority return service_tier as priority when Fast serves the request.
Can Fast mode fall back to Standard?
Yes. OpenAI documents that rapid traffic ramps can downgrade some Fast requests to default speeds and standard rates; the response then reports service_tier as default.
Official sources
Source check: August 19, 2026. Recheck supported models, tier values, prices, discounts, ramp rules, rate limits, access, capacity, region, fallback, and SLA terms.