OpenAI released the Agents API on September 10, 2026, saying it provides a powerful harness for developers to build and run cloud agents. It is the company's first major software update since the launch of Codex.
OpenAI reported an evaluation score improvement from 0.71 to 0.85, measured on subagent workflow efficiency. That compares with the previous setup, which had higher latency and cumbersome orchestration.
The Agents API is built on the Codex harness and targets enterprise workflows requiring long-running, reliable agents. Availability begins with a public beta, initially for all developers.
"With the Agents API, our evaluation score went from 0.71 to 0.85. The subagent support in the API is great and drastically sped up our workflow," said Jack Weissenberger, CTO, Ciridae. The new APIs gave their team a 4x latency reduction.
The announcement follows the scaling of Codex and ChatGPT for Work to millions of users. OpenAI said the API enables developers to focus on unique workflows while the company manages the infrastructure.
OpenAI did not say how the API will scale with future model updates, and raised the question of how developers will optimize for subagent performance. The company said it will iterate based on user feedback during the public beta.
Source: openai
