OpenClaw has switched its default runtime to Codex for agents built on OpenAI's technology. The move is designed to boost performance and streamline how those agents handle tools, the company said.
A new default for AI agent operations
By adopting Codex as the standard runtime, OpenClaw is changing the underlying engine that runs its OpenAI-based agents. The company described the integration as a direct effort to improve two areas: raw performance and the way agents interact with external tools. Tool handling has been a pain point for many agent platforms, often requiring custom workarounds that slow down deployment. Codex aims to solve that with a more unified approach.
Performance and tool handling gains
OpenClaw didn't release specific benchmark numbers, but pointed to the overall aim of making agents faster and more reliable when calling functions or accessing data. The runtime change means developers using OpenClaw will get those improvements without having to modify their agent code — the upgrade happens at the platform level. For teams already running agents in production, the swap should be transparent, the company indicated.
Setting standards for multi-model platforms
The integration also carries implications beyond OpenClaw's own user base. In its announcement, the company framed the decision as a step toward setting new standards for multi-model AI platforms. By making Codex the default, OpenClaw is betting that a single, optimized runtime can serve as a foundation for agents that might pull from different large language models underneath. That approach could push other platform providers to rethink their own default runtimes or tool-handling strategies.
OpenClaw has not disclosed whether it plans to extend Codex support to agents built on other model providers, like Anthropic or Google. For now, the change applies specifically to OpenAI agents on its platform.
