The model, built upon open-weights MiniMax H3, currently holds the top spot on independent benchmarks from Design Arena and Artificial Analysis. Design Arena awarded the model an Elo rating of 1,341 on its Image-to-Video leaderboard, placing it ahead of competitors such as FLUX.3 and Gemini Omni Flash. According to Artificial Analysis, the model achieved an Elo rating of 1,201, outperforming systems like ByteDance’s Seedance 2.0 and Kling 3.0.
In section Releases
Fal Debuts H3 Max, Challenging Video Generation Speed and Quality
San Francisco-based generative media platform fal has released H3 Max, a post-trained video model capable of generating five-second clips in approximately three seconds. By co-designing the inference engine alongside the model architecture, the company claims to have bypassed the traditional trade-off between high-fidelity visual output and processing speed.
Batuhan Taskaya, head of engineering at fal, attributed the performance to a unified development approach. By optimizing the system stack during the training phase rather than as an afterthought, the team achieved throughput rates roughly 35 times higher than the official MiniMax H3 endpoint. The model is accessible via the fal API, Playground, and Agent, with a promotional pricing period offering 50% discounts on generation costs through September 7, 2026.
Comments (0)
No comments yet. Be the first!