Chinese artificial intelligence firm MiniMax has introduced H3, its latest multimodal video generation model, aimed at challenging the dominance of closed-source competitors through an open-weight approach and competitive pricing. The launch comes as rival ByteDance unveils its newest video model, Seedance 2.5.
According to benchmarking platform Artificial Analysis, H3 currently stands as one of the most powerful AI models globally for video editing. However, it ranks below Google’s Gemini Omni Flash in text-to-video tasks, and trails both ByteDance’s Seedance 2.0 and Gemini Omni Flash in image-to-video performance.
H3 is intended for commercial use across various industries including film, advertising, video gaming, and product design. It is accessible via an application programming interface (API) and MiniMax’s consumer platform, Hailuo AI. The company has adopted a Community Licence model, which allows free non-commercial use and permits commercial usage for organizations with annual revenues under US$20 million, subject to attribution. This licensing approach reflects a broader trend in the Chinese AI sector as companies seek to commercialize large-scale models while supporting open-source communities, according to analysis from Goldman Sachs.
Artificial Analysis noted that once publicly downloadable, MiniMax’s H3 may become the most powerful open-weight video AI model available globally. The firm is also pursuing an aggressive pricing strategy amid stiff competition. At 768p resolution (1,366 by 768 pixels), H3 is approximately half the price of comparable models, and at 2K resolution (around 2,048 by 1,080 or 2,560 by 1,440 pixels), it costs roughly a third. Jefferies analysts reported that H3 is priced at about US$7.80 per minute of generated video at 2K resolution with audio, compared to ByteDance’s Seedance 2.0, which costs US$22.45 per minute for 1080p video. Notably, MiniMax’s offering is costlier than Google’s Gemini Omni Flash, priced at US$56 per minute.
ByteDance concurrently released Seedance 2.5, promoting its capability to produce 30-second video clips in a single run with enhanced editing features. The company highlighted the model’s potential applications in education and industrial manufacturing sectors. Meanwhile, ByteDance has intensified its focus on AI-driven enterprise solutions, integrating the development team behind its office productivity tool Lark with the team managing its Doubao AI chatbot.
Industry observers report that ByteDance’s annual recurring revenue from large language models, which estimates yearly income based on current subscription levels, has recently reached approximately US$4 billion, underscoring the company’s expanding footprint in the AI market.