Codex makes fewer bugs, but more people actually use Claude
The 'State of AI Coding 2026' report from cubic.dev compares and finds that code written with OpenAI's Codex has a lower bug rate than other tools. Yet Claude is the tool more actually use day to day. The core observation is that code quality and real-world adoption don't necessarily move together.
Key points
- cubic.dev published a 'State of AI Coding 2026' report comparing
- Code from OpenAI's Codex reportedly has a lower bug rate than
- Claude has a larger user base / higher actual adoption
- Suggests code quality and popularity don't necessarily correlate