I tracked Claude Code vs Codex usage across 13 heavy users for a month. It's basically a tie, and most people use both

A study tracked AI code usage among 13 heavy users for a month, analyzing approximately 1 billion tokens. The findings indicate a near tie between Claude Code and Codex, with most users employing both. The top single models observed were opus 5.5 (22%) and gpt-5.6-sol (20%), followed by opus 5 and gpt-6-astra. This suggests a diverse and flexible approach to AI coding tools among experienced developers.