Fixed
Improved reliability of long-running streaming requests
Streaming requests that involve long prompt processing, or that run while the service is handling other large requests, are now significantly more reliable. Previously such requests could fail with a connection timeout before any output was returned.
If you saw intermittent timeouts on large-context requests, they should now complete normally.
No changes are required in your code. Streaming responses may now include standard SSE comment lines during processing pauses; conformant clients, including the official OpenAI SDKs, ignore these automatically. See Streaming if you maintain a custom SSE parser.