How HappyHorse‑1.0 Surpassed Seedance 2.0 in AI Video Generation Rankings

Artificial Analysis, a well‑known AI evaluation platform, uses an Elo‑based blind‑test mechanism where users vote between two models without knowing their identities, ensuring reliable rankings for generative video models.

Within a week of Seedance 2.0’s debut, the mysterious HappyHorse‑1.0 entered the leaderboard and immediately claimed the top spot in both the text‑to‑video and image‑to‑video categories, pushing Seedance 2.0 to second place. In the text‑to‑video track HappyHorse‑1.0 leads by roughly 59 points, and in the image‑to‑video track by about 35 points. Users describe the gap as a “break‑through,” noting that the combined score difference from ranks 2 to 10 barely reaches the margin between HappyHorse‑1.0 and Seedance 2.0.

Sample outputs illustrate the model’s exceptional realism: a middle‑aged woman’s facial expressions, skin texture, and tears appear natural; an elderly chef’s interaction with hot soup shows precise skin pores, oil sheen, and physically accurate motion. Viewers on X (formerly Twitter) repeatedly exclaimed that these results are stronger than Seedance 2.0.

Artificial Analysis compared HappyHorse‑1.0 against Seedance 2.0, Kling 3.0 Pro, grok‑video‑imagine, and PixVerse V6, highlighting HappyHorse‑1.0’s superior visual fidelity, adherence to physical laws, and multilingual support.

Rumors attribute the model to Alibaba’s “Future Life Lab.” The narrative cites Zhang Di, former Vice President of Technology at Kuaishou who led the Keling AI project, released Keling 2.0 in April 2025, and rejoined Alibaba in November 2025 to focus on multimodal innovation. The team’s low‑key approach aligns with the anonymous launch.

The community response has been intense: X users praised the model’s quality, joked about a “black‑horse” for the 2026 Year of the Horse, and raised concerns about the limited sample size and whether the model can maintain its ranking over time.