MiniMax released H3 on July 31, adding a rare combination to the video-generation race: a commercial API price and a stated plan to open the model weights. The release shifts the question from demo quality alone to a harder one: if 2K audiovisual generation is priced by the second, how long can closed video models keep a premium?
H3 accepts text, images, video and audio in one context. It generates up to 2K resolution, 15 seconds, with native stereo sound. MiniMax points to advertising, branding, e-commerce, product design, UI/UX and games, with emphasis on instruction following, text and brand rendering, and V2V motion transfer.
The open-weight promise is not the file yet
“We plan to open up the model weights in the coming days, subject to applicable laws and regulations.”
Developers still need the license, commercial-use terms, hardware guidance and a technical report before treating H3 as a deployable open model.
Pricing makes the story sharper
Chinese financial outlets reported H3 at RMB 0.8 per second for 2K generation, about one-third of comparable flagship video models. MiniMax says 768p is under half of mainstream 720p pricing.
At RMB 0.8 per second, a 15-second 2K clip costs roughly RMB 12 before platform packaging. That figure puts H3 into a repeatable iteration budget rather than a showcase-only tier.
Unified context is the technical bet
MiniMax describes H3 through Contextual Omni Representation, H3-VAE, H3-Omni Transformer and In-Context Regeneration. Video, audio, images and text are handled as one creative context instead of separate tasks.
The company says most source material requires about 100,000 tokens of inference before being distilled to roughly 4,000 tokens on average. H3-VAE provides a 4x gain in effective sequence length, and workload separation lifted training throughput by nearly 30%.
The next checks are concrete: whether weights appear, whether the license works for companies, and whether the community can reproduce usable text, motion and audio control on non-official hardware.
Sources: MiniMax official research blog, Shanghai Securities News, The Beijing News, CocoLoop, 21st Century Business Herald; MiniMax site used to verify 2K/15-second output, native stereo audio, open-weight commitment, RMB 0.8-per-second pricing, 768p comparison and nearly 30% training-throughput claim.