15d (edited) • News
The Guy Who Built Kling Just Beat Kling
A mystery AI video model called HappyHorse-1.0 showed up on the Artificial Analysis benchmark last week with no name attached. Within days it was #1 in text-to-video AND image-to-video, beating Seedance 2.0, Runway Gen-4.5, and every other model on the board.
Then Alibaba raised their hand. Turns out it was built by their ATH AI unit, led by Zhang Di, the former VP of Kuaishou who built Kling AI's technology. The guy who built the previous champion just built the new one. For a different company.
The numbers aren't close. 1333 Elo in text-to-video (60 points ahead of #2). 1392 Elo in image-to-video (37 points clear). And here's the part that matters: it generates video and audio together in a single pass. Not two separate models stitched together. One transformer, 40 layers, everything at once.
They've confirmed it's going open source. API access starts April 30.
If that open source release actually delivers benchmark-level quality, the math changes for everyone paying monthly for Runway or Kling. The best model in the world, free to download and run locally.
Worth paying attention to.
What do you think this means for the paid tools? Does free + best quality kill the subscription model? Drop your take below.
Founding Members are getting a full technical breakdown of HappyHorse's architecture and what it means for your workflows this week.
7
6 comments
The Guy Who Built Kling Just Beat Kling
skool.com/aiography
Hollywood craft meets creative AI. Learn how to generate studio-quality content, secure clients, and get paid. From someone who's actually made films.
Leaderboard (30-day)
1
🔥
+15
2
+6
3
+5
4
+3
5
+3
Powered by