Open Responses: New open inference standard built on the Responses API for agents

AI Impact Summary

Open Responses introduces an open, standardized inference format designed to replace the Chat Completion workflow for agentic workloads. It extends the Responses API with explicit fields for raw reasoning, encrypted_content, and summaries, and models streaming as semantic events, enabling richer visibility into multi-step reasoning and tool invocation pipelines. By separating Model Providers from Routers and supporting both internal and external tools, it enables coordinated multi-provider inference and more predictable orchestration across vendors. The change signals a shift toward interoperable, end-to-end agent loops, which will require updates to clients, routers, and provider implementations, but promises more consistent capabilities and observability once adopted.