China's MiniMax H3 is the first open model to top an AI video ranking
MiniMax released the weights for H3, a 33-billion-parameter model that ranks first in video editing, second in text-to-video, and third in image-to-video on Artificial Analysis. The opening is real but partial — the 2K module and the H3-Context-IR prompt component stay closed, local ComfyUI runs cap at 768p, and commercial use is limited to companies under $20 million in revenue. Within those limits the weights are fine-tunable on custom footage, characters, or a house visual style, putting a downloadable model in the same tier as closed systems like ByteDance's Seedance 2.5, released the same day.
Source: the-decoder.com ↗
The open weights do allow fine-tuning on custom footage, characters, or a specific visual style.
Artificial Analysis / MiniMax model card
Why this matters
- → First open-source model ranks top in any major AI video benchmark
- → Fine-tunable weights enable custom styles and characters locally
- → Narrows gap between open and closed video generation systems
Open beats closed