MiniMax releases H3 video model to challenge rivals
Chinese AI startup MiniMax has released its H3 video-generation model, which can process text, images, video and audio and create clips of up to 15 seconds in 2K resolution with native stereo sound. The Shanghai firm says H3 targets commercial use in advertising, e-commerce, product design and games.
Key Takeaways
- MiniMax released H3 on July 31, 2026, intensifying competition with ByteDance and Kuaishou.
- H3 can edit content and transfer movements across mixed input formats, with native stereo audio.
- MiniMax plans to release H3 model weights within days and designed the system for Chinese-made chips.
- The company claims 2K video generation costs less than one-third of mainstream rival products.
According to Reuters, MiniMax stepped up competition in China's fast-growing video-model market with Friday's launch. For related coverage, visit our Streaming & TV Alerts hub.
What can MiniMax H3 actually do?
H3 is positioned as a multimodal video-generation model. It can ingest text, images, video and audio, then produce clips of up to 15 seconds at 2K resolution with native stereo sound.
MiniMax also says the model can edit existing content and transfer movements from one video to another. Users can supply instructions and reference material in different formats, which matters for remixing footage rather than generating clips from scratch alone.
Why does the H3 release matter for creators?
Chinese video-model competition has heated up since ByteDance released Seedance 2.0 earlier this year and Kuaishou launched Kling 3.0. MiniMax is pitching H3 at commercial applications including advertising, e-commerce, product design and games.
Cost is part of the pitch. MiniMax said generating 2K video would cost less than one-third of mainstream rival products, a claim aimed at brands and studios testing AI video at scale.
Founded in 2022, MiniMax is one of China's so-called "AI tigers," a group of well-funded startups competing with domestic tech giants and U.S. leaders. It became the second of that group to list publicly when it floated in Hong Kong in January, following Z.AI.
Will MiniMax open the H3 model weights?
MiniMax said it plans to release H3's model weights within days, allowing users to download and customise the underlying system. Reuters notes that move would extend the open-weight approach increasingly used by Chinese AI developers into video generation, where many prominent models remain proprietary.
The company also said H3 was designed to work with several Chinese-made chips, amid a broader push in China's AI sector to reduce dependence on U.S. semiconductors.
How were investors framing MiniMax before H3?
Days before the launch, a Goldman Sachs interpretation of a MiniMax founder meeting, summarized on Moomoo, cast H3 as a multimodal catalyst beyond video alone. Goldman reiterated a Buy rating with a 12-month target of HK$860, citing H3 among factors that could narrow MiniMax's valuation discount.
Earlier buzz also built around a cryptic MiniMax teaser campaign. Techgenyz reported the company posted "It's coming" with a #MiniMaxH3 video repeating "the world is multimodal," after the Speech 2.8 voice update and alongside Hailuo AI branding.
Reuters had previously reported in early July, citing a person with direct knowledge, that MiniMax planned to launch H3 later that month and was developing a 2.7-trillion-parameter language model. Friday's H3 release puts the video piece of that roadmap into the market.