OpenAI Codex merges 345K PRs in 35 days, dwarfing GitHub Copilot and other coding agents
Jun 20, 2025
Key Points
- OpenAI's Codex merged 345,000 GitHub pull requests in 35 days, vastly outpacing GitHub Copilot's 10,000 and other coding agents in the same window.
- Codex's scale advantage stems from OpenAI's installed base: millions of ChatGPT users can frictionlessly switch to the coding tool without adoption friction.
- The PR count inflates from low-rigor contributions like documentation tweaks, making the technical rigor and actual impact of Codex's merged work unclear.
Summary
OpenAI's Codex merged 345,000 pull requests on GitHub in 35 days. GitHub Copilot sits at 10,000 merged PRs, Devon at nearly 20,000, Cursor Agents at 1,000, and Codegen at 1,600. Codex also shows the highest success rate among the cohort.
The scale difference appears driven primarily by distribution rather than product superiority alone. OpenAI Teams already has broad workplace penetration, giving Codex an installed base advantage. Millions of software engineers already use ChatGPT, making it frictionless to switch and test Codex as an alternative.
Not all PRs are created equal. A PR can range from meaningful feature work to trivial documentation tweaks or minor fixes to high-profile projects, the kind of low-friction contribution that inflates contributor counts on popular repositories. The variety and actual technical rigor of these 345,000 PRs remains unclear from the data alone.
The raw volume and success rate still signal that OpenAI Codex is gaining material adoption velocity in the developer tool market. GitHub Copilot, despite generating roughly half a billion dollars in revenue, is being outpaced on this particular metric by a product that launched more recently and benefits from OpenAI's broader user base.