MiniMax H3 has pushed open AI video models into a new position: first place in a major video ranking category. MiniMax has released the H3 video model weights, and Artificial Analysis ranks the model first in Video Editing, second in Text-to-Video, and third in Image-to-Video.
The release matters because it puts an open model at the top of a video ranking for the first time. It also shows how quickly AI video systems are moving beyond simple text prompts into workflows that combine images, video, audio, and editing instructions.
What MiniMax H3 can take in and produce
H3 is a 33-billion-parameter model built to process several kinds of media together. According to the source, it can work with text, images, video, and audio as part of the same generation process.
The output is also multimodal. H3 can generate four- to 15-second clips with stereo sound, placing it in the group of AI video models that treat audio as part of the video result rather than as a separate afterthought.
The model card gives a clearer view of the input range. A single prompt can include up to nine reference images, three video clips, and three audio clips. That means the model is not limited to a short text instruction when users want to guide the result.
For creators and developers, that input flexibility is important. Reference images can help define a subject or visual direction, while video and audio clips can give the model more context for motion, timing, or sound. The source does not describe every workflow this enables, but the supported inputs point toward more controlled AI video generation than text alone.
Why the ranking result stands out
Artificial Analysis ranks H3 first in Video Editing. It also places the model second in Text-to-Video and third in Image-to-Video. Those rankings put H3 near the front across several core AI video tasks, not only in one narrow use case.
The most notable point is the open-model angle. The source states that H3 is the first open model to top an AI video ranking. In a field where many leading systems are closed, releasing weights changes what some users can do with the model.
Open weights can make a model more adaptable. In this case, the H3 release allows fine-tuning on custom footage, characters, or a specific visual style. That is a meaningful distinction from systems where users can only submit prompts through a closed service.
Still, open weights do not mean the full MiniMax video stack has been released. H3 arrives with important omissions that shape how it can be used outside MiniMax systems.
The open release has limits
Two parts remain closed. The 2K resolution module is not included, and H3-Context-IR is also not part of the release.
H3-Context-IR is described as the component that translates prompts and reference material into a structured intermediate format. Without it, users running the model outside MiniMax need to prepare context themselves.
MiniMax has published prompting guides for that context preparation. The source does not present those guides in detail, but it makes clear that local users will need to handle this step instead of relying on the closed H3-Context-IR component.
Resolution is another practical constraint. Running H3 locally in ComfyUI tops out at 768p. The excluded 2K resolution module means local use does not match the complete system described around MiniMax H3.
- Included: H3 video model weights.
- Not included: the 2K resolution module.
- Not included: H3-Context-IR.
- Local limit: H3 in ComfyUI tops out at 768p.
Commercial use comes with a revenue condition
The license also matters. Commercial use is only permitted for companies making under $20 million in revenue.
That condition means H3 is not an unrestricted open release for every business. Smaller companies under that threshold may have commercial options, while companies above it would need to account for the license limitation.
For non-commercial experimentation, research, or internal testing, the release still gives users access to weights they can study and adapt. But any production plan has to begin with the license terms, especially if the work is intended for paid products or client-facing services.
Competition is moving at the same time
MiniMax was not the only company moving in AI video on the same day. ByteDance released its closed Seedance 2.5 the same day, and that model generates 30-second clips with built-in audio.
The contrast is clear from the source: MiniMax H3 is an open-weight release with strong ranking results and local-use constraints, while Seedance 2.5 is closed and supports longer clips. The source does not give a direct ranking comparison between those two models, so the safest takeaway is that both releases point to active competition in AI video generation.
For now, H3's importance comes from where it lands in the rankings and what MiniMax has chosen to release. It gives open-model users access to a leading AI video system, while keeping some parts of the full pipeline closed. That combination makes H3 both a milestone and a reminder that openness in AI video can still come with significant technical and licensing boundaries.