OpenAI Responses API adds async tools, steering
OpenAI’s Responses API now lets GPT-6 Astra continue independent work while application-managed tools run asynchronously. Developers can also steer active responses over WebSocket while preserving completed work and task context.
This pushes tool-calling agents toward event-driven runtimes, reducing idle latency while making application orchestration more complex.
- –Async tools keep long-running calls from blocking unrelated reasoning or output
- –Developers must manage pending work, retries, correlation, and original call IDs
- –Mid-turn steering enables corrections without restarting an entire agent task
- –Steering requires GPT-6 Astra and Responses API WebSocket transport
- –The latency gains are compelling, but Astra’s premium token pricing makes efficient orchestration essential
DISCOVERED
1h ago
2026-09-04
PUBLISHED
1h ago
2026-09-04
RELEVANCE
AUTHOR
OpenAI