MiniMax H3 Ranks First in AI Video Editing Evaluation as the First Open-Weight Model
MiniMax H3 is a 33-billion-parameter open-weight video model generating 4–15 second clips with stereo sound and ranking first in video editing at release.
Model MiniMax H3
Developer MiniMax
Release 2026-08-03
Type Open-weight video generation model
A 33-billion-parameter multimodal video model
MiniMax has released the weights for its video generation model H3. In Artificial Analysis evaluations, it ranked first in the video editing category—the first open-weight model to do so—while ranking second in text-to-video and third in image-to-video. These rankings reflect the evaluation results at the time of release and may change later.
H3 has 33 billion parameters and processes text, images, video, and audio together. It generates stereo-sound videos ranging from 4 to 15 seconds long. According to the model card, a single prompt can include up to nine reference images, three videos, and three audio files.
The release has some limitations. The 2K-resolution module and H3-Context-IR, which converts prompts and reference materials into a structured intermediate format, were not released. Local execution through ComfyUI is limited to 768p, and users must prepare the context themselves according to the publicly available prompt guide. However, the released weights can be fine-tuned for original videos, characters, or specific styles.
Under the license, commercial use is permitted only for companies with less than $20 million in revenue. We corrected the previous post's mistaken classification as a general “AI news briefing” and reorganized the model's input range, generation length, unreleased components, and commercial-use terms based on the original source.