Alibaba's Happy Horse Ranks #1 on Artificial Analysis But Fails Tests
Alibaba's Happy Horse video generation model ranks approximately 100 points above Seedance 2.0 on the Artificial Analysis video leaderboard but breaks physics adherence and prompt fidelity in independent real-world test scenarios. Available for free on Alibaba's platform, independent reviewers ran direct head-to-head comparisons with Seedance 2.0 on princess and zoom-shot prompts and found Seedance 2.0 clearly superior. The gap between leaderboard ranking and real-world performance raises benchmark contamination as a plausible explanation.
Why It Matters
Leaderboard-first model releases are becoming a reliable anti-pattern in 2026; treating self-reported Artificial Analysis rankings as marketing rather than ground truth is now the default stance for practitioners.