fal launches H3 Max video generation model

The generative media platform fal has released H3 Max, a new post-trained video generation model built upon the open-weights MiniMax H3 model. Optimized by fal’s inference team, H3 Max is designed for faster-than-real-time generation, capable of producing a five-second video in approximately three seconds. This represents a significant increase in throughput compared to the official MiniMax H3 endpoint.

According to fal, the model achieves high quality through post-training that improves prompt adherence and visual fidelity while utilizing an inference engine specifically designed for the model. H3 Max has achieved top rankings on independent AI model benchmarks, including Design Arena’s Image-to-Video leaderboard, where it outperformed the official MiniMax H3 and other leading models like Seedance 2.5 and Gemini Omni Flash.

In addition to its speed, the model offers competitive pricing. For example, fal provides 768p generation at a rate of approximately $0.08 per second, making it a cost-effective option for developers and enterprises compared to some existing industry alternatives.