China's AI: MiniMax H3, Seedance 2.5, DeepSeek V4 Launch

Today marks a busy day for China's domestic large AI models, with MiniMax H3, Seedance 2.5, and DeepSeek V4-Flash all being released. MiniMax launched H3, its first open-source multimodal generation model. It integrates video generation, editing, and multimodal understanding into one model. H3 supports various inputs like text, images, audio, and video, and can output 2K resolution content, generating up to 15-second audiovisual videos. Its pricing is competitive at 0.8 yuan per second for 2K videos. ByteDance also rolled out Seedance 2.5, a new-generation video creation model. This model extends single video generation to 30 seconds. Meanwhile, DeepSeek announced the public beta launch of its V4-Flash official version API. This version includes key upgrades in Agent and coding capabilities. What's interesting is that two of these models focus on video creation, while the other targets developer coding terminals. This shows a significant push in diverse AI applications from domestic manufacturers.