unknown

A hands-on comparison of three leading AI coding models (GPT-5.2-Codex high, Claude Opus 4.5, and Gemini 3 Pro) tested on two real-world coding tasks in an 8K+ star repository. Claude Opus 4.5 delivered the most consistent and polished results with fastest completion times but highest cost ($0.94-$1.78). GPT-5.2-Codex high produced excellent code quality with strong architecture but required 20-26 minutes per task due to high reasoning mode. Gemini 3 Pro was most efficient and cost-effective with heavy cache utilization but produced more minimal, less polished implementations. Both tasks involved building complex features: a global action palette with keyboard navigation and a usage analytics dashboard with data persistence.