OpenAI Launches Public Beta of Agents API, Enabling Developers to Build AI Agents

OpenAI has launched the public beta of its Agents API, providing developers with the framework and infrastructure for Codex and the enterprise version of ChatGPT. Developers can build production-grade AI agents through the API, with costs based on token consumption and tool invocation. The Agents API emphasizes long-running and complex task execution capabilities, allowing developers to specify tasks, models, tools, and runtime environments in a single API call to create agents. Core capabilities include automatic context compression, parallel tool invocation, and multi-agent collaboration, supporting agents to run continuously for hours or even days. Customer cases show that after integrating SafetyKit, the cost per case decreased by 60%, Hypha reduced the agent response failure rate by 86%, and Cirridae's evaluation score improved from 0.71 to 0.85, with latency reduced by four times. The Agents API operates in the same sandbox environment as Codex and ChatGPT, and has integrated partners such as Blaxel, Cloudflare, DigitalOcean, Oracle, and Vercel, supporting various CPU, GPU, memory, and deployment environments. Built on the open-source Codex execution framework, the Agents API supports tools like MCP, custom functions, and web search, marking OpenAI's expansion of Codex into a developer-focused agent infrastructure platform.