Same prompt in Gemini 4 Pro and GPT-6 Astra yields two very different Three.js butterflies
The author gave the same prompt to Gemini 4 Pro and GPT-6 Astra to build an interactive 3D mechanical butterfly in Three.js, and the two results differed completely in UI, structure and rendering.
I built the same thing twice, once with GPT-6 Astra, once with Gemini 4 Pro. An interactive 3D mechanical butterfly in Three.js. One prompt, copy-pasted into both. I changed nothing else. What came back were two completely different machines - different UI, different structure, different rendering. Gemini 4 Pro gave me "Papilio Mechanica": a clockwork wireframe, exploded schematic, spec panel, drag-to-orbit controls. GPT-6 Astra gave me "Papilio Nocturne": a gyroscopic heart, iridescent azure-and-gold wings, a hidden helical transmission, exterior/interior modes. 99% of people run one model and assume that's what the task looks like. The 1% run the same prompt across models and ship the better build. That's not a prompting gap. That's a stack gap, and it takes one evening to close. Prompt → Astra → Gemini → Compare → Ship the winner This one test taught me more than any $1000 AI course.