OpenAI Agents API launch
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
OpenAI launched the public beta of its Agents API on September 10, opening Codex's agent infrastructure.
The move pushes OpenAI deeper into commercializing agent infrastructure developers build on.
Background
OpenAI is opening the core agent architecture that powers its flagship coding assistant Codex to external developers, a step in commercializing its AI agent infrastructure. The API is a managed service built on the open-source Codex harness, with OpenAI handling session management, orchestration, and context management so developers can focus on their own agents. It is free to use, with pricing based on token consumption and tool usage. Alongside it, OpenAI began offering API access to GPT-Live-1, its real-time voice model first previewed in ChatGPT in July 2026.
How it unfolded
- Jul 2026OpenAI previews GPT-Live-1, its full-duplex real-time voice model, in the ChatGPT interface.
- Sep 10, 2026OpenAI launches the public beta of its Agents API for all developers, offering the Codex harness as a REST API.
- Sep 10, 2026OpenAI also begins offering API access to GPT-Live-1 and launches a new Data agent in ChatGPT Work.
Who’s saying what
- Official
- OpenAI says scaling Codex and ChatGPT for Work showed long-running agents need a harness that manages context, uses tools efficiently, and coordinates subagents, plus infrastructure that keeps them running reliably for days.
- Caution
- The API currently has US-only data residency and no zero-data-retention option, which the material notes limits regulated workloads for now.
Still unverified
The claim that the Agents API is 'free to use' appears alongside statements that pricing is based on token and tool usage and container time; the material does not fully reconcile these. Early customer metrics (60% cost reduction for SafetyKit, 86% fewer failures for Hypha, 4x faster latency for Cirridae) are single-source figures not independently confirmed.