AI Showdown: Codex and Claude Code Leads Clash Publicly

The conflict began on July 12 when Theo observed that GPT-5.6 Sol performed better in Claude Code's environment than in Codex for some tasks. Tibo (Thibault Sottiaux), OpenAI's Codex lead, then publicly shared a method to use GPT-5.6 Sol within Claude Code via a local proxy, joking he would reset user quotas if the trick got him banned. In August, developer Alex Getman followed the method and his Anthropic account was suspended for 'suspicious signals'. Boris Cherny, Claude Code's creator, responded by first inviting Tibo to work at Anthropic, then stating Anthropic does not ban users for using other models in its harness; he believed a different account classifier triggered the ban and later confirmed the account was restored. Tibo declined the job offer, reiterated his belief in 'freedom of harness', and reset usage quotas for all ChatGPT Work and Codex paid users, despite many having just refreshed their weekly quotas. CEO Sam Altman joked on X, praising Tibo. The incident underscores that models and agent harnesses are being decoupled, with users increasingly choosing their preferred model-harness combinations, and both OpenAI and Anthropic are positioning their harnesses as independent layers.