According to Watchful AI Monitoring, MINIMAX-W(00100.HK) has officially launched the MiniMaxM3 model, achieving state-of-the-art performance in tasks such as programming and agent intelligence. Utilizing the new Attention architecture MSA (MiniMaxSparseAttention), it can handle up to 1M tokens for long-range context. MiniMaxM3 is also a native multimodal model supporting both image and video inputs, with the capability to interact with a computer desktop.