Hong Kong-listed artificial intelligence startup MiniMax reported a sharp market reaction after unveiling H3, a new open-source multimodal generation model designed to handle unified context across text, images, video and audio. The company said H3 can generate videos with native stereo sound up to 15 seconds in length at 2K resolution.
Shares of MiniMax surged 17.4% to HK$230.0 by 04:24 GMT following the announcement. The company positioned H3 as a general-purpose multimodal model intended for commercial deployment in sectors including advertising, e-commerce, gaming, product design and content creation.
MiniMax highlighted H3's cost efficiency when compared on a per-second basis with mainstream competing models. The firm said 2K video generation with H3 costs less than one-third of comparable models, while at 768p resolution pricing is stated to be less than half that of models generating 720p video.
To support developers and broaden compatibility across AI hardware platforms, MiniMax indicated it plans to release the model's weights in the coming days, subject to applicable laws and regulations.
MiniMax described several technical innovations underpinning H3, naming Contextual Omni Representation, the H3-Omni Transformer and In-Context Regeneration as key components. According to the company, these technologies enable improved instruction following, enhanced multimodal understanding and high-resolution content generation.
The launch reflects a broader push by Chinese AI companies to deploy increasingly capable open-source models. MiniMax framed H3 as offering industry-leading price-performance and as targeting a range of commercial use cases where multimodal generation is relevant.
Summary
MiniMax unveiled H3, an open-source multimodal AI model that processes unified context across text, images, video and audio and can generate up to 15 seconds of 2K video with native stereo sound. The company's Hong Kong-listed shares jumped 17.4% to HK$230.0 by 04:24 GMT. MiniMax emphasized H3's price-performance advantages versus comparable mainstream models and said it will release the model's weights in the coming days, subject to laws and regulations.
Key points
- H3 is a general-purpose multimodal generation model supporting text, images, video and audio, with 2K video output up to 15 seconds and native stereo sound.
- MiniMax reported significant cost advantages for H3: 2K video generation at less than one-third the per-second cost of comparable mainstream models, and pricing at 768p that is less than half versus models generating 720p.
- The model targets commercial applications across advertising, e-commerce, gaming, product design and content creation, and MiniMax plans to release model weights subject to legal and regulatory considerations to aid developer adoption and hardware compatibility.
Risks and uncertainties
- Regulatory and legal constraints may affect the timing and scope of the planned release of H3's model weights, which MiniMax said is subject to applicable laws and regulations - this impacts developers and hardware compatibility efforts.
- Claims about relative price-performance are comparative in nature; MiniMax framed H3 as costing less than one-third or less than half of competing models at specified resolutions, but these statements imply dependency on how "comparable mainstream models" are defined and priced.
- The announcement comes amid a competitive landscape of Chinese AI firms releasing open-source models to challenge international rivals, creating uncertainty about market responses and competitive dynamics.