Analysis
MiniMax released H3, its newest multimodal video generation model, capable of producing 2K resolution clips up to 15 seconds long with native stereo sound -- a technical bar few competitors currently clear at MiniMax's stated price point. The model also supports video editing and motion transfer, letting users apply one video's movement to a different reference image or prompt using mixed-format inputs.
MiniMax says H3 costs less than one-third per second of generated video compared with mainstream competitors, an aggressive pricing position in a category where rendering costs have historically been a major barrier to consumer and mid-market adoption. Full model weights are scheduled for open-source release in the coming days, a deliberate strategic choice given that both of MiniMax's most direct rivals -- ByteDance's Seedance and most closed Western video labs -- keep their systems proprietary.
“MiniMax's bet is that open weights plus aggressive pricing wins developer mindshare even if H3 doesn't lead on every quality benchmark against closed competitors.”
The launch lands in the same week ByteDance rolled out its own upgraded Seedance 2.5 model, intensifying a Chinese AI-video race that is increasingly being fought on two axes simultaneously: raw generation quality and openness. MiniMax's bet is that open weights plus aggressive pricing wins developer mindshare even if H3 doesn't lead on every quality benchmark against closed competitors.
For investors tracking the generative media landscape, MiniMax open-sourcing a genuinely competitive video model is a meaningful data point in the broader trend of Chinese AI labs using openness as a distribution strategy against better-funded but closed Western competitors -- the same playbook DeepSeek has run in text models. What to watch: how quickly Western video-generation startups respond on price once H3's weights are actually available for self-hosting, and whether open-source video generation follows the same rapid commoditization curve that open-source text models did.
