Sonnet 5, Fable 5, GPT-5.5, and Gemini compared on Three.js coding benchmark

A Reddit user tested four models (Sonnet 5, Fable 5, GPT-5.5, and Gemini) on generating a cyberpunk alley in Three.js from scratch with the same prompt. The one-shot benchmark uses a locked 10-second camera dolly to isolate code quality.