The post-trained H3 Max model generates 5 second 720p video clips in under 3 seconds for $0.04 per second of 768p resolution. fal Research built the model on top of the open weight MiniMax H3, resulting in a system that now ranks first in Image to Video and third in Text to Video on Artificial Analysis leaderboards. The release is nearly 50x faster than the base H3 model, according to fal.

The model was co-optimized with fal's in-house inference engine to achieve an Image to Video generation time of 6.4 seconds, which is 18x faster than the average. fal Research intends to release the weights for H3 Max, a move that would establish it as the highest ranked open weights model on both the Artificial Analysis and Film Arena boards.