OpenAI Codex Lead's Claude Code Setup Tip Triggers Account Ban, Declines Anthropic Invite

On July 12, 2026, developer Theo noted in a social media discussion of GPT-5.6 Sol that the same model outperformed Codex on some tasks when run within the Claude Code environment. Thibault Sottiaux, OpenAI's Codex lead, subsequently published a configuration method for connecting GPT-5.6 Sol to Claude Code, saying the setup could be completed in roughly five minutes. The approach does not modify Claude Code itself: developers retain the official CLI, and a locally running CLIProxyAPI forwards model requests to other model providers. Claude Code's tools and permission system remain unchanged, with only the underlying reasoning model swapped for GPT-5.6 Sol or similar models.

In August, developer Alex Getman configured the integration as described and his Anthropic account was suspended, with the system citing "suspicious signals" as the reason. Getman publicized the incident on GitHub and filed an appeal, noting he was not certain whether the proxy configuration directly caused the ban. Sottiaux offered assistance in the comments, noting he does not work at Anthropic, and asked whether other users had encountered similar issues. Boris Cherny, head of Claude Code at Anthropic, responded that Anthropic does not ban accounts for using other models within their own harness, and said the initial assessment was that a separate account classifier had been triggered, with the team investigating. Cherny added that Claude Code supports other models, which can be integrated through proxy solutions such as LiteLLM.

Soon afterward, Cherny updated that Getman's account had been restored and that the team was taking measures to prevent a recurrence.

Sottiaux reiterated that users should have the freedom to choose their models and declined Cherny's job offer. He also fulfilled an earlier promise by resetting usage quotas once for all ChatGPT Work and Codex paid subscribers. Some users noted that most weekly quotas had already refreshed normally on August 8, limiting the practical value of the reset. Sottiaux responded that another reset would take place the following Monday. OpenAI's Sam Altman made a brief comment on the matter on social media.

The technical context of the incident includes OpenAI's description of the harness used by Codex and ChatGPT Work as a Rust orchestration system connecting models, tools and user environments, managing context, tool invocations and the agent loop. Anthropic's official documentation defines Claude Code's LLM Gateway Protocol and provides configuration instructions for connecting third-party gateways. During the episode, developer Theo's observation that the same model performs differently across harnesses drew attention, though the discussion did not alter existing product definitions.