TechFlow News, June 1: MINIMAX-W (00100.HK) announced the launch of its MiniMaxM3 model. According to the announcement, “MiniMaxM3 achieves state-of-the-art performance on specialized tasks such as programming and agent-based operations. It employs a newly proposed attention architecture—MSA (MiniMax Sparse Attention)—supporting context lengths of up to 1 million tokens. As widely anticipated, it is also a native multimodal model capable of processing image and video inputs and interacting with desktop environments.
These three capabilities are essential for cutting-edge closed-source models overseas. M3 is the first domestic model to integrate all these features—and currently the only open-source model offering them.” (Jinshi)
Add to Favorites
Share to Social Media
