2026-10-05
Kling 3.0 holds the top spot today, and blind arena voting backs this up with real separation: it leads the video generation leaderboard with a score of 1934, visibly ahead of the next closest competitor, built on the multi-shot sequences introduced earlier this year that keep subject consistency intact across different camera angles for up to 15 seconds. Veo 3.1 holds second on audio-video sync that's still unmatched in this ranking, perfectly timed character dialogue that makes it the pick whenever a project needs tightly synced, dialogue-driven scenes. Runway Gen-4.5 and Seedance 2.0 tie for third, each winning on a different strength: Runway on camera control and generative editing for production pipelines, Seedance on multi-shot scene generation with native lip sync and improved physics. I'm nudging Alibaba's HappyHorse 1.0 up slightly today, from 8.8 to 8.9, after this week's arena data placed it second overall at 1816, ahead of Seedance 2.0 Fast's 1747, a genuinely strong showing for a model that only recently arrived from the Alibaba team. Hailuo 02 and Luma's Ray3 continue to hold the mid-tier on solid but unremarkable performance across every factor in this ranking. Nothing else moved this week.
Kling 3.0's Multi-Shot Consistency Is the Real Edge
A 1934 arena score comes directly from subject consistency holding across different camera angles for up to 15-second sequences, a genuine first in this category.
Veo 3.1 Still Owns Dialogue-Driven Scenes
Perfectly synced character dialogue remains unmatched here, the clear pick whenever a project needs talking, not just motion.
Runway and Seedance Tie on Different Strengths
One wins on camera control for production pipelines, the other on native lip sync and physics, both earning the same third-place score.
HappyHorse Earns a Bump on Fresh Arena Data
A second-place 1816 arena score, ahead of Seedance 2.0 Fast, is a genuinely strong result for Alibaba's newest entrant, worth a small nudge up today.