Codex Merged into ChatGPT: Feature Upgrades Overshadowed by 'Token Assassin' Backlash
OpenAI integrated its standalone Codex programming tool into ChatGPT desktop app on July 9, 2026, but the launch faced immediate criticism over...
On July 9, 2026, OpenAI merged its standalone Codex programming tool into the ChatGPT desktop app, billing it as an 'AI super-app' combining coding, chat, and work modules. However, the rollout quickly triggered backlash: users reported token consumption gaps of 5-10x between medium and high reasoning tiers, a hidden basic chat entry in the sidebar, and even an incident where GPT-5.6 Sol allegedly deleted nearly all files on a Mac user’s device.
The product’s form shifted drastically: from a standalone local coding app for paid professional developers (focused on single-repo editing), Codex now sits as one of three core modules (Chat, Work, Codex) in the unified ChatGPT desktop app, available to free users on macOS and Windows. While existing Codex users retained their projects and settings, the new UI folded the familiar chat interface into the sidebar, leaving many non-programmers struggling to find basic chat functions.
Toolchain integration saw a major upgrade: via the Model Context Protocol (MCP), the new Codex connects deeply with JetBrains IDEs (PyCharm, IntelliJ IDEA) for project-level configuration and approval workflows, Azure DevOps for PR summaries and pipeline fault localization, and Oracle AI Agent Studio for proxy app development—far beyond the old version’s single-repo limitations.
Feature-wise, Codex expanded to full-cycle engineering management: inline annotation for code/Markdown, sidebar PR reviews (reducing tool switching), cross-repo project management, and GPT-5.6-driven Computer Use (handling code, browser, and desktop operations with permission checks). But the GPT-5.6 upgrade (Sol/Terra/Luna tiers) came at a cost: even the medium Sol tier consumed more tokens than the old Codex’s highest tier.
Reputation was split: Google AI’s Logan Kilpatrick called it a 'super-app era milestone,' while tech blogger M.G. Siegler labeled it 'terrible.' Hacker News users吐槽ed confusing UI and token costs. On July 17, OpenAI’s product lead Thibault Sottiaux responded with fixes: improved sidebar switching, cross-device history sync, and reassurance that Codex’s core experience remained intact.
Who should use it? Enterprise teams needing cross-repo collaboration, automated PR reviews, or Azure DevOps integration will find it optimal—its cross-repo management and PR workflows outshine the old standalone version. Individual developers or casual users should wait: high token costs and non-programmer-unfriendly UI make Cursor or Claude Code better alternatives.
Sources
- Toutiao Article: https://m.toutiao.com/article/7664902458707100223/