Claude Opus 5.5 leads AI video generation benchmark

Artificial Analysis has released the Code2Video-AA benchmark, a new evaluation method that ranks AI models based on their ability to create marketing videos from code. In the inaugural leaderboard of 27 models, Claude Opus 5.5 secured the top position with an Elo rating of 1536, outperforming competitors across all scoring axes, including engagement, prompt intent, composition, temporal consistency, and craft.

Claude Sonnet 5.5 followed in second place, while GPT-6 Astra took third. The benchmark utilizes a unique process where models do not paint pixels directly; instead, they write code to build HyperFrames HTML compositions and animate them using the GSAP JavaScript library. The evaluation relies on blind pairwise comparisons by human judges to determine the most effective outputs based on 114 different product and brand marketing briefs.

Results indicated a significant performance gap between proprietary and open-weight models, with leading open-weight models scoring approximately 480 points lower than the top-tier proprietary models.