OpenAI adds service tier restrictions for API projects
OpenAI added controls for permitted API service tiers on October 21, 2026. Its release note says project settings can select which tiers a project may use, while organization settings can restrict tiers across every project.
These controls concern processing modes. A separate October 6 change reorganized API usage tiers into Build, Launch and Grow, with upgrades based on credit purchases. That earlier account classification change is separate from the new service tier restrictions.
Which requests are covered
The linked error guidance covers default, flex and priority. It evaluates fast as priority. Scale Tier remains outside this project policy. A project cannot enable a tier blocked by its organization.
Automatic selection can still fail
A disallowed requested or resolved tier returns HTTP 400 with type invalid_request_error and error.param set to service_tier. Omitting that parameter or choosing auto can also fail if the resulting tier is disallowed. Automatic selection therefore does not guarantee an allowed route.
This is distinct from HTTP 429 errors for exhausted credits, rate limits or spend limits. The documented remedy is an allowed request tier or a permitted settings change.
Processing defaults still matter
OpenAI’s Fast mode FAQ describes existing configuration options. A project can default to Fast mode, with individual requests able to select a processing mode. The FAQ dates the Priority processing rename to July 30.
Fast mode carries a token price premium over Standard processing. For monitoring, the FAQ directs users to group usage by Service Tier and cost by Line Item. These views distinguish processing choices from the amount spent.
Important details remain undocumented
The release note and error guide do not specify editing permissions, initial restriction presets or effects on requests already running. Their policy scope also does not explain Ultrafast handling.
ByteForward reviewed published documentation without changing account settings or running live API tests.
Related coverage examines OpenAI’s earlier Fast mode introduction and pricing changes.
Illustrative archival photograph of MareNostrum 4 at Barcelona Supercomputing Center in 2017. Photo by Gemmaribasmaspoch via Wikimedia Commons, Creative Commons Attribution ShareAlike 4.0.
