fal Releases H3 Max, a Post-Trained MiniMax H3 Video Model That Generates Five Seconds in About Three
- fal H3 Max generates a five-second video in about three seconds of wall time, which fal says is roughly 35 times the throughput of MiniMax's official H3 endpoint.
- Design Arena ranks H3 Max first on its recent image-to-video leaderboard at 1,341 Elo, ahead of official MiniMax H3 at 1,333.
- Artificial Analysis ranks fal's H3 model first on its image-to-video-with-audio leaderboard, scoring 1,201 Elo across 2,177 samples.
- fal post-trained the open-weights MiniMax H3 with new data and its in-house reinforcement learning framework, using human preference studies for quality, prompt understanding, and aesthetics.
- fal co-designed H3 Max with its inference engine, retaining speed optimizations only when evaluations preserved the model's quality gains.