Introducing H3 Max, new post-trained video model by fal Research....
fal@fal
138 views
Aug 29, 2026
~1 min read
Advertisement
1
Introducing H3 Max, new post-trained video model by fal Research.
H3 Max ranks #1 for overall quality, prompt understanding, and aesthetics against leading video models, on both first-party and third-party independent evaluations while generating a 5-second 720p video under 3 seconds.
H3 Max is 50% off for the next week, making it the highest quality, fastest and cheapest model for overall video generation tasks.
H3 Max ranks #1 for overall quality, prompt understanding, and aesthetics against leading video models, on both first-party and third-party independent evaluations while generating a 5-second 720p video under 3 seconds.
H3 Max is 50% off for the next week, making it the highest quality, fastest and cheapest model for overall video generation tasks.
2
We benchmarked H3 Max against 12 leading video models.
It ranked #1 across all three dimensions we measured: overall quality, prompt understanding, and aesthetics, winning the majority of matchups against every model we tested.
It ranked #1 across all three dimensions we measured: overall quality, prompt understanding, and aesthetics, winning the majority of matchups against every model we tested.
3
H3 Max is also fast.
It generates a 5-second video under 3 seconds, which is roughly 35x the throughput of the official H3 endpoint. It's faster than every other model in our comparison by a wide margin.
It generates a 5-second video under 3 seconds, which is roughly 35x the throughput of the official H3 endpoint. It's faster than every other model in our comparison by a wide margin.
4
H3 Max is post-trained by fal on top of the open-weight base MiniMax H3 model.
We introduced significant new data during post-training, tuning specifically for stronger prompt adherence and aesthetics. And spent a huge portion of our post-training compute on verifiable RL tasks.
We introduced significant new data during post-training, tuning specifically for stronger prompt adherence and aesthetics. And spent a huge portion of our post-training compute on verifiable RL tasks.
5
We also designed the architecture of H3 Max around fal’s in-house inference engine where we spent the past 4 years optimizing diffusion models with it.
We prioritized quality first, then pushed speed as far as we could without compromising the aesthetics and prompt understanding.
The result is a model that runs significantly faster without the usual tradeoff in quality.
We prioritized quality first, then pushed speed as far as we could without compromising the aesthetics and prompt understanding.
The result is a model that runs significantly faster without the usual tradeoff in quality.
6
H3 Max is available now on fal, 50% off this week.
Try it now:
fal.ai/models/minimax…
fal.ai/models/minimax…
Try it now:
fal.ai/models/minimax…
fal.ai/models/minimax…
7
Correction:
The above graphic reflects inaccurate Artificial Analysis rankings for some of the models. The current rankings are:
Gemini Omni Flash: #4
Wan 3.0: Unranked
Seedance 2.5: Unranked
FLUX.3: Unranked
Kling 3 Pro: #14
The above graphic reflects inaccurate Artificial Analysis rankings for some of the models. The current rankings are:
Gemini Omni Flash: #4
Wan 3.0: Unranked
Seedance 2.5: Unranked
FLUX.3: Unranked
Kling 3 Pro: #14
