Opus 5.5 vs GPT 6 Astra in 3D web benchmark
The prompt is itself a reusable template for stress-testing frontier models on long-form, constraint-heavy front-end generation — explicit bans on lazy shortcuts like background-image parallax make differences in instruction-following easy to see. Practitioners building with Claude or GPT for complex single-shot web apps can copy the prompt verbatim to compare how each model handles architectural detail, lighting and code completeness. Both model pages are linked for direct inspection.
