MiniMax officially launches the M3 model: coding and agent tasks reach cutting-edge levels
MiniMax has rolled out the MiniMax M3 model, featuring the brand-new MSA (MiniMax Sparse Attention) architecture, supporting up to 1M ultra-long context. This multimodal model can handle image and video inputs as well as desktop operations. It achieves cutting-edge performance in specialized tasks like coding and agents.
Why it matters: The 1M context and multimodal capabilities of MiniMax M3 signify a continued breakthrough by Chinese AI teams in the general AI space, with the MSA sparse attention architecture providing a new technical pathway for long-context reasoning.
#MiniMax #AI #人工智能 #大模型 #Web3