Claude, GPT-6 Lead New Code-to-Video Benchmarks as Models Master Animation Without Diffusion

AI models are generating high-quality animated videos by writing HTML, CSS, and JavaScript code rather than using traditional video diffusion models—and a new wave of benchmarks is measuring who does it best. Claude Opus 5.5 dominates the Code2Video-AA evaluation with an Elo score of 1536±26, outpacing GPT-6 Astra by 109 points on marketing video design tasks. The shift reflects a fundamental realization: large language models naturally excel at generating code, making code-based video rendering a more tractable path than pixel-level diffusion for structured, deterministic content.