MiniMax H3 launch lifts MINIMAX-W shares 12.86% in Hong Kong

MINIMAX-W shares jumped 12.86 percent to HKD 230 after the Hong Kong-listed AI company unveiled MiniMax H3, an omni-modal generative model that produces video with native dual-channel audio at up to 15-second 2K resolution.

The company said the model supports unified understanding of text, images, videos, and audio inputs, and plans to open the model weights within days, subject to compliance with relevant laws and regulations. Trading volume reached 4.269 million shares with turnover of HKD 998 million at midday, while short selling accounted for $81.91 million, or 8.207 percent of total turnover.

The launch places MiniMax in direct competition with OpenAI's GPT-4o and Google's Gemini, both of which already support multimodal input and output. By opening model weights, MiniMax is following the open-source strategy of Meta's Llama and Alibaba's Qwen, betting that developer adoption will offset the absence of a proprietary platform.

Open-weight strategy targets developer adoption

The decision to release model weights — a departure from the closed-source approach of OpenAI and Google — could accelerate adoption among developers who want to deploy the model on their own infrastructure. Meta's Llama series demonstrated this playbook, becoming the most-downloaded open-weight model family since its 2023 debut. Alibaba's Qwen has pursued a similar path in China, with its Qwen2.5 models ranking among the top open-weight releases on Hugging Face.

MiniMax's dual-channel audio generation is a differentiator. Most current multimodal models, including GPT-4o and Gemini, can generate audio but typically produce a single audio track. Native dual-channel sound enables spatial audio output, which could matter for video production, gaming, and immersive media applications. The 15-second 2K video output also exceeds the typical output length of several competing models, though MiniMax did not disclose the test conditions for these capabilities.

The shift toward omni-modal models reflects a broader industry trend. OpenAI's GPT-4o, released in May 2024, was among the first to natively handle text, image, and audio inputs in a single model. Google's Gemini 1.5 Pro extended this to video understanding. MiniMax's H3 goes further by generating video and audio natively, rather than stitching together separate models for each modality.

The company's decision to open weights within days — rather than months after launch — reflects confidence in the model's technical foundation. However, regulatory compliance in China adds uncertainty. The company said the release is "subject to compliance with relevant laws and regulations," which could delay or restrict availability in certain jurisdictions.

The 12.86 percent single-day surge reflects investor enthusiasm for MiniMax's omni-modal capabilities, but the company faces a crowded field. In China alone, Alibaba's Qwen, Baidu's Ernie, and ByteDance's Doubao all compete for developer mindshare. The open-weight strategy could help MiniMax differentiate, but it also means the company must monetize through enterprise services and API usage rather than proprietary model access.

MINIMAX-W trades on the Hong Kong Stock Exchange under ticker 00100.HK. The company has not disclosed revenue or valuation figures tied to the H3 launch. The broader AI model market in China is intensifying as domestic players race to match Western capabilities while navigating export controls on advanced chips. MiniMax's omni-modal approach — combining text, image, video, and audio understanding with generation — could give it a niche in media and entertainment applications, where dual-channel audio and high-resolution video output are critical requirements.

The stock's midday gain of HKD 26.20 per share came on turnover of HKD 998 million, suggesting active institutional participation. Short selling at 8.207 percent of total turnover indicates some investors are betting against the rally, which could create volatility in the near term. The company's next milestone will be the actual release of model weights, which could either confirm or temper the market's enthusiasm depending on how the open-source community responds.

This article is for informational purposes only and does not constitute investment advice.