MiniMax unveils its own open-source LLM with industry-leading 4M token context

MiniMax has released the MiniMax-01 series, a new family of open-source AI models designed for ultra-long context handling and enhanced AI agent development. The series includes MiniMax-Text-01, a large language model with a remarkable 4 million token context window, and MiniMax-VL-01, a visual multimodal model.

These models utilize a novel Lightning Attention mechanism, reducing computational complexity and improving efficiency. MiniMax-01 rivals top-tier models in performance, particularly in long-context evaluations, while offering significantly lower API costs. The company aims to support the growing need for sustained memory and efficient communication in AI agents, and encourages collaboration with developers and researchers.