fal unveils H3 Max, a faster post-trained version of the MiniMax H3 video model

At a glance

fal released H3 Max, a video model made by further training the openly available MiniMax H3 with new data and tuning it for maximum speed. fal says it ranked first in its own human preference tests for overall quality, prompt understanding and aesthetics. It produces a 5-second video in under 3 seconds.

Key points

1. 1

   It renders a 5-second clip in under 3 seconds, roughly 35 times the throughput of the official MiniMax H3 endpoint.

2. 2

   fal says it is on average 15 times faster than other models of similar quality.

3. 3

   Human raters compared it with 12 video models, including Gemini Omni Flash, Wan 3.0, Seedance 2.5, Kling 3 and Veo 3.1.

4. 4

   It was trained and served on NVIDIA GB200 NVL72 systems, and speedups that hurt its quality ranking were dropped.

Why it matters

It breaks the usual tradeoff in video AI where higher quality means slower output, showing high-quality clips can be made in less time than they last.