MiniMax M3

MiniMax develops proprietary LLMs across various modalities, including a trillion-parameter MoE model, a speech model with low latency and native support for major Asian languages, and a state-of-the-art text-to-speech and text-to-video models. The company's flagshig MiniMax M3 is a frontier open-weight AI model optimized for coding, agentic workflows, and long-context reasoning. Powered by the proprietary MiniMax Sparse Attention (MSA) architecture, it supports a context window of up to 1 million tokens and native multimodality for text, image, and video understanding.