Orient Securities: MiniMax H3 Releases AI Video Track to Welcome Open Source

The Zhitong Finance App learned that Orient Securities released a research report saying that MINIMAX-W (00100) released a new video model H3 after a lapse of more than half a year to complete native multi-modality, same audio and video output, and reference generation. The difference is that the final film has excellent visual effects and is cost-effective. It is expected that scenes such as skits, advertising, visual design, and special effects will accelerate penetration. The bank believes that attention should be paid to vertical multi-modal AI application opportunities, and is optimistic that technological breakthroughs and cost optimization will accelerate the development of industry trends, driving user growth, payment penetration, and commercialization to the next level. Furthermore, particular attention is paid to companies that are leading in multi-modal technology and are commercialized in a closed loop.

Orient Securities's main views are as follows:

MiniMax released a new video model H3, which is cost-effective and has a differentiated visual style, which is expected to open up new downstream scenes

Following the release of Hailuo 2.3 on 25q4, MiniMax officially released the new video model H3 at the end of July '26. This update fixes features such as native multi-modality, native dual-channel audio and video output, and all-round reference. By default, 2K resolution is provided, which has significantly improved character and style consistency, text retention, and prompt compliance. Currently ranked 2/3/1 on Artificial Analysis' Wensheng video/graphic video/video editor list.

Extremely cost-effective, expected to benefit cost-sensitive scenes such as skits

The H3 supports 2K resolution by default, and thanks to multiple technologies such as Contextual Omni Representation, H3-VAE, H3-Omni Transformer, and Incontext Regeneration, H3 provides excellent cost performance. According to the bank's statistics, the H3 2K unit price is 0.8 yuan/second. Compared with the SD2.0 VIP 1080P is 3 yuan/second, the SD2.5 720P is 2.5 yuan/second, and the Keling 3.0 1080P is 1.2 yuan/second. In the AI content category, scenarios such as mass production and more emphasis on creative efficiency and production costs, such as skits, are more beneficial. Furthermore, due to H3's unique stylized expressiveness, the bank's AI video tools are expected to accelerate penetration in scenarios such as brand advertising, product design, UI/UX, and game trailers.

Take the lead in launching the video circuit, the leading model open source, and focus on the development of the open source ecosystem in the industry

Unlike LLM, the current SOTA video model is still mainly closed source. Previously, the open source camp included models such as Alibaba Wanxiang, Tencent Hybrid Video, and LTX, but there is still a gap compared to closed source models such as Seedance and Keling. The new H3 model chose open source while retaining the capabilities of leading models. The bank believes that further attention should be paid to the opening up of subsequent models. For enterprises, control of deployment, data and workflow can be obtained, and material asset creation and utilization can be completed within the enterprise after privatization; for the industry as a whole, video models can also be customized and expanded, and have better customization capabilities and developer ecology, thus accelerating the overall penetration of the AI video industry.

Risk Alerts

AI technology iteration falls short of expectations; AI application implementation falls short of expectations; AI commercialization falls short of expectations