OpenAI lets one API project choose regional processing per request

On August 21, 2026, OpenAI added per-request regional processing for API customers. A project configured with Global geography can use the same API key against a region-prefixed API domain to request processing in a supported region, rather than requiring separate region-specific projects and credentials for every workload. OpenAI’s documentation shows US and Europe routing examples and says existing eligibility, data-retention controls, endpoint support and model-support requirements still apply.

Multi-region SaaS products often have to route customer workloads according to contractual or regulatory geography. Moving region selection from project configuration into the request path can simplify credential management and shared infrastructure, especially when one service handles customers in multiple jurisdictions. The trade-off is that residency policy now becomes application logic: builders must reliably select the correct base URL, prevent fallback into the wrong processing region, and understand that regional processing support is not the same as regional data storage everywhere.

OpenAI now allows a project whose geography is Global to send an individual API request through a region-prefixed domain while using the project’s existing API key. OpenAI documents examples for US and European processing. That removes one source of operational duplication for products that previously would have needed separate projects or credentials primarily to control request geography.

The data-controls documentation distinguishes regional storage from regional processing. The US and Europe support regional processing for eligible workloads, while several other listed geographies support storage residency without regional model processing. Builders should therefore check the current support matrix rather than assuming a storage-eligible region also supports processing there.

Per-request routing does not bypass OpenAI’s existing data-residency requirements. Non-US regional controls can require approved abuse-monitoring arrangements and contractual retention terms, and eligible models released on or after March 5, 2026 carry a 10% regional-processing uplift according to OpenAI’s current documentation. Endpoint and model restrictions also continue to apply.

The operational benefit comes with a governance obligation: the same client and key can now reach the default global endpoint or an eligible regional endpoint. Teams need explicit tenant-to-region rules, tests that prevent accidental global fallback, and observability that records which processing domain handled a request. Regional processing also applies to Customer Content rather than every category of system or account data, so it should not be described as blanket infrastructure residency.