OpenAI Opens Up Codex Agent Underlying Capabilities to Developers, Agents API Enters Public Beta

Beating AI News Flash: OpenAI has officially launched the public beta of the Agents API, opening up to all developers the agent framework and infrastructure previously used to power Codex and enterprise ChatGPT. Developers can now build production-grade AI agents directly through the API, with fees billed by Token consumption and tool invocation volume.

The Agents API focuses on long-running and complex task execution capabilities. Developers can create an agent by specifying the task, model, tools, and runtime environment in a single API call. Its core capabilities include automatic context compression, parallel tool invocation, and multi-agent collaboration, enabling agents to run continuously for hours or even days.

Customer cases disclosed by OpenAI show that after integrating SafetyKit, the cost per case dropped 60%; Hypha reduced agent response failure rates by 86%; and Cirridae's evaluation score improved from 0.71 to 0.85, while latency was reduced by 4x.

On the infrastructure side, the Agents API can use the same sandbox environment as Codex and ChatGPT, and has already integrated partners including Blaxel, Cloudflare, DigitalOcean, Oracle, and Vercel, supporting different CPU, GPU, memory, and deployment environments.

OpenAI also stated that the Agents API is built on the open-source Codex execution framework and supports tools such as MCP, custom functions, and web search. This opening means OpenAI is further expanding Codex from an AI coding product into an agent infrastructure platform for developers.