MiniMax teases H3 as a unified multimodal model for Hailuo AI
MiniMax is framing H3 as a single system for text, images, video and sound, aimed at carrying creative context across media.
By Ryan Merket · Published
Why it matters
MiniMax is pitching H3 as a unified system for text, images, video and sound. If it can carry shared context across those formats, H3 could reduce the tool handoffs involved in producing longer, more complex creative work.
Yan Junjie's MiniMax teased H3, saying only "It's coming." and tagging Hailuo AI, MiniMax's video-generation product. The wording ties H3 to Hailuo's creator-facing service, though it does not establish a product launch.
Yan, MiniMax's founder, chairman, CEO and CTO, came to foundation models through research rather than consumer software. He earned a mathematics degree from Southeast University, completed a doctorate in artificial intelligence at the Chinese Academy of Sciences and conducted postdoctoral research at Tsinghua University. Before starting MiniMax, he spent more than six years at SenseTime, including as a vice president and vice-head of its research institute.
That background fits MiniMax's work across video generation, character apps and foundation models.
H3 is a workflow bet
MiniMax had already shown the direction before the X teaser. A July 17 report from Shanghai Securities News carried by Sina Finance said MiniMax previewed H3 at WAIC 2026 as a next-generation multimodal generation model. The exhibit described a system that could interpret a shared context made from text, images, video and sound, then produce a more coherent result across those formats.
If MiniMax delivers that design, H3 would reduce the handoffs that define many current creative workflows. A user often moves among separate tools for story development, image generation, video generation, audio and editing. MiniMax is pitching a model that understands the project as one continuous context.
That is a different product goal from competing solely on the quality of a short generated clip. Runway's Gen-4.5, for example, emphasizes motion quality, prompt adherence, temporal consistency and fine-grained controls. H3 will still face those tests. MiniMax is also trying to move up a layer, from generating an asset to managing more of the creation process around it.
Useful comparisons will require details on output length and resolution, synchronized audio, character and scene consistency, editing controls, latency and generation cost. API availability will show whether MiniMax sees H3 mainly as a Hailuo feature or as infrastructure that other product teams can build into their own software.
Yan is pairing product cadence with financing plans
H3 follows MiniMax M3, which MiniMax released on June 1 as an open-weight model aimed at coding, agents and multimodal input. MiniMax says M3 supports a context window of up to 1 million tokens and can process image and video inputs. The release established the other half of Yan's product structure: the M-series for language, coding and agents, and the Hailuo line for visual creation.
The release sequence also follows a large capital plan. MarketScreener reported that MiniMax sought to raise about HK$16.04 billion, or about $2.05 billion, through a share sale and convertible bond issue in July.
If completed, that financing would help fund model training, inference capacity, research hiring and distribution while MiniMax competes with better-known video platforms. That capital plan does not validate H3's technical performance.
MiniMax's economics still make model efficiency and monetization central. MiniMax reported $79.0 million in 2025 revenue and a $250.9 million adjusted net loss. Its year-end accounts listed $1.05 billion across cash, deposits and several categories of financial assets. MiniMax also says its products have served more than 300 million individual users across more than 200 countries and more than 1 million enterprises and developers, while its video models have generated more than 600 million videos.
MiniMax's annual report divides revenue between AI-native products and open-platform and other AI-based enterprise services; the revenue table does not break out Hailuo AI as a standalone line. H3 therefore has to advance MiniMax's model work while giving creators another reason to pay MiniMax directly.
The launch details will decide the story
MiniMax's prospectus says MiniMax has been committed to developing multimodal models "from day one." In its 2025 results, MiniMax separately quoted Yan as saying MiniMax had built full-modality research and development capabilities during the year. H3 narrows MiniMax's strategy into a practical wager: creators will prefer a system that carries intent across text, visuals, motion and sound instead of forcing them to rebuild context at every step.
The next announcement needs to turn that wager into a product. A model card, public access terms and reproducible demonstrations would show how much of H3 is a unified architecture and how much is orchestration around specialized components. Pricing will reveal whether MiniMax plans to use its new capital to undercut rivals, protect margins or reserve H3 for higher-value production work.
The teaser is thin by design. MiniMax has pursued substantial new financing, built distribution through Hailuo and described where it believes multimodal creation is heading. H3 now has to connect those pieces.