OpenAI has officially opened public beta access to its Agents API for external developers. This API supports the agent frameworks for Codex and the enterprise version of ChatGPT, delivered via programming interfaces, with billing based solely on token usage and tool call volume. Core capabilities include automatic context compression, parallel tool invocation, and multi-agent collaboration, enabling the creation of production-grade agents in a single API call.
OpenAI is opening up the core agent architecture that powers its flagship coding assistant, Codex, to external developers, marking a significant step in the commercialization of AI agent infrastructure.
On Wednesday, OpenAI announced that the Agents API has entered public beta and is now available to all developers.
The API makes the agent framework and infrastructure, previously used exclusively to support Codex and the enterprise version of ChatGPT, available via programming interfaces. Developers can access it without additional fees, with billing based solely on actual token consumption and tool invocation volume.
Data disclosed by early adopters indicates significant improvements in cost efficiency, operational effectiveness, and reliability.
For instance, after migrating its case review workflow to the Agents API, SafetyKit reduced per-case processing costs by 60%. Hypha lowered agent response failure rates by 86% by decoupling the agent execution framework from sandbox environments. Meanwhile, tech firm Cirridae reported an increase in evaluation scores from 0.71 to 0.85, alongside a fourfold reduction in latency.
Create production-grade agents with a single API call
The core design philosophy of the Agents API is to lower the engineering barrier for building long-running agents.
Developers can create production-ready agents by specifying tasks, models, tools, and runtime environments in a single API call, eliminating the need to build underlying infrastructure such as context management and sub-agent orchestration.
The API offers three key capabilities:
First, automatic context compression. As sessions approach the context window limit, the system automatically compresses earlier content, enabling agents to operate continuously for hours or even days;
Second, tool search and parallel invocation. Relevant tool definitions are loaded on demand to reduce token consumption, while supporting the parallel execution of multiple tool calls.
Third, multi-agent collaboration. The primary agent can decompose complex tasks and distribute them to multiple concurrently running sub-agents. Each sub-agent maintains its own context independently, with the primary agent ultimately aggregating the results.
Nash, a logistics technology company, stated that its platform has deployed thousands of long-running AI agents via the Agents API to manage hundreds of millions of global delivery tasks. These agents operate over cycles ranging from several hours to several days.
Flexible environment selection with ecosystem partners covering diverse needs.
At the computing environment level, the Agents API offers multiple deployment options.
OpenAI provides self-hosted sandboxes that share the same infrastructure as Codex and ChatGPT, enabling rapid startup and elastic scaling.
(The Agents API provides AI agents with the same framework and infrastructure as Codex.)
Meanwhile, OpenAI has established first-tier integration agreements with ecosystem partners including Blaxel, Cloudflare, Daytona, DigitalOcean, E2B, Modal, Oracle, Runloop, and Vercel. These partnerships provide developers with differentiated options covering private cloud deployments, specific storage mechanisms, and varying CPU, GPU, and memory configurations.
This design enables the Agents API to address the full spectrum of needs, from rapid validation by startup teams to enterprise-grade production deployments.
Long Lake, an AI platform in the construction and home services sector, stated that leveraging this API allows its team to deploy agents across industries within hours, rather than the several days previously required.
Open-source frameworks and versioned updates reduce iteration costs.
The Agents API is built on the open-source Codex execution framework, allowing developers to review its public codebase to understand the core logic of model invocation, tool orchestration, and context management.
OpenAI is responsible for maintaining and continuously optimizing this framework, enabling developers to automatically benefit from each model upgrade without needing to refactor their agent architectures.
OpenAI stated that the Agents API is currently in public beta and will undergo rapid iterations based on developer feedback before progressing to an official release.
The company also noted support for built-in tools such as the MCP protocol, custom functions, and web search. For developers, no additional subscription fees are required for the API itself; this pricing strategy helps lower adoption barriers while directly linking revenue sources to actual usage volume.
Editor/Stephen