Source: OpenAI Streaming API Responses Guide
OpenAI API guide on streaming the Responses API with typed semantic SSE events, and the moderation risk of partial output.
Contents
- What: official API documentation page
- Author/publisher: OpenAI
- Date: living document; read 2026-11-14
- Original: Streaming API responses
Summary
With stream=true, the Responses API emits typed semantic events instead of raw token chunks. The guide lists the lifecycle, output-item, content-part, text, refusal and function-argument events, recommends a minimal subset for text streaming, and warns that partial output is harder to moderate.
Key claims
- Lifecycle events:
response.created,response.in_progress,response.completed,response.failed,error. - Structure events:
response.output_item.added/done,response.content_part.added/done. - Content events:
response.output_text.delta/done,response.output_text.annotation.added,response.refusal.delta/done,response.function_call_arguments.delta/done; plus tool-specific events (file search, code interpreter). - For plain text, listening to
response.created,response.output_text.delta,response.completedanderroris enough. - Streaming makes moderation harder: partial completions are harder to evaluate, and moderation results arrive only after full generation.