OpenAI Agents API Launches in Public Beta for All Developers

OpenAI released the Agents API in public beta to all developers on September 10, 2026. The API exposes the open-source Codex harness, which coordinates model calls, tools, and context, through a single interface. OpenAI runs the agent loop on its own infrastructure; developers can alternatively use their own infrastructure or one of nine partner sandboxes. Built-in capabilities include automatic context compaction for multi-hour runs, tool search to reduce token usage, programmatic tool calling with parallel execution, and multi-agent support that lets a main agent delegate to parallel subagents. There are no additional API fees; developers pay for tokens, tools, and container time. Each agent loop turn incurs charges for the underlying model's token usage, hosted tools such as web search and file search carry per-call fees, and Code Interpreter adds container-time charges. Current limitations include US-only data residency and no Zero Data Retention support, restricting use for regulated workloads.