MiniMax launches H3, a multimodal video model with native audio, stock jumps 17%

MiniMax released H3 on Friday, a new multimodal generation model that processes text, images, video and audio in a single unified context and outputs native 2K video with built-in stereo sound, in clips up to 15 seconds long. The Chinese AI company said it plans to open-source the model's weights within days, letting developers download and customize the underlying system rather than relying solely on an API.

The launch matters because it pushes MiniMax's open-weight strategy into video, a domain still dominated by closed, expensive systems from OpenAI (Sora) and Google (Veo). MiniMax has built its reputation over the past year on shipping models that land close to proprietary frontier benchmarks while charging a fraction of what US labs charge and releasing the weights publicly. H3 applies that same playbook to video: MiniMax says generating 2K footage on H3 costs roughly a third of what mainstream rival products charge, a pricing gap the company is positioning as a direct competitive lever against Sora and Veo for commercial video work.

Technically, H3 introduces three components MiniMax calls Contextual Omni Representation, the H3-Omni Transformer, and In-Context Regeneration. Together they let the model accept mixed inputs — a text prompt alongside a reference image, video clip, or audio track — and hold them in a shared representation space, which MiniMax says improves instruction-following and multimodal understanding over its prior Hailuo video line. The model also targets compatibility with domestic Chinese AI chips, part of a broader push by Chinese AI firms to reduce dependence on US-made semiconductors for both training and inference.

MiniMax is pitching H3 squarely at commercial use cases: advertising, e-commerce product visuals, product design, and games — markets where 15-second, ready-to-use clips with synced audio are more valuable than longer cinematic output. That's a narrower target than Sora or Veo's broader creative-generation pitch, but it maps directly to where studios currently spend the most on stock and produced video content.

The market reaction was immediate: MiniMax Group's Hong Kong-listed shares (HKG: 0100) jumped as much as 17.4% following the announcement, before settling to a reported 15.1% gain on the day. Some of that move is tied to a separate catalyst — rising investor expectations that MiniMax is nearing inclusion in the Stock Connect program, which would let mainland Chinese investors buy the stock directly through the southbound trading link and could bring tens of billions of Hong Kong dollars in fresh inflows. The H3 launch appears to have amplified that existing momentum rather than created it from scratch.

For the AI video market, H3's planned open-weight release is the detail worth watching most closely. If MiniMax follows through within days as promised, H3 would become one of the first frontier-tier multimodal video models with publicly available weights, giving developers and smaller studios a self-hostable alternative to the closed APIs that currently define the category — as first reported by Reuters, with additional detail from Investing.com and KFGO.