MiniMax H3 outperforms Tencent HunyuanVideo 1.5 in video generation benchmarks

MiniMax just grabbed the crown in open-source video generation. The company’s new H3 model, released around July 31, has climbed to the number one position among open video models, dethroning Tencent’s HunyuanVideo-1.5 by a margin that’s hard to ignore: approximately 280 points higher in the Video Arena benchmark across both text-to-video and image-to-video categories.

For context, HunyuanVideo-1.5 was only released in November 2025 and had been sitting comfortably near the top of the open-model rankings.

What makes H3 different

H3 generates video clips up to 15 seconds long at 2K resolution with native stereo audio output. H3 also supports what MiniMax calls a unified input format, meaning it can accept text, images, video, and audio as inputs. You can feed it a photo and a text prompt, or hand it an existing video clip and ask for edits.

The model ships with open weights, which means developers can download it, modify it, and run it locally.

Community adoption happened fast

Within days of H3’s release, the developer community had already integrated it into ComfyUI, a popular node-based workflow tool used by AI artists and video creators. GGUF quantization, a technique that compresses models to run on consumer-grade hardware rather than expensive data center GPUs, followed shortly after.

HunyuanVideo-1.5, with its 8.3 billion parameters, had established a decent foothold in this same community. But the performance gap H3 has opened up creates a strong gravitational pull.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.