{"id":13430,"date":"2025-06-25T11:59:16","date_gmt":"2025-06-25T11:59:16","guid":{"rendered":"https:\/\/www.europesays.com\/us\/13430\/"},"modified":"2025-06-25T11:59:16","modified_gmt":"2025-06-25T11:59:16","slug":"minimax-releases-m1-a-456b-hybrid-attention-model-for-long-context-reasoning-and-software-tasks","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/us\/13430\/","title":{"rendered":"MiniMax Releases M1: A 456B Hybrid-Attention Model for Long-Context Reasoning and Software Tasks"},"content":{"rendered":"

MiniMax has introduced MiniMax-M1<\/a>, an open-weight language model designed for long-context reasoning and tool use. Based on the earlier MiniMax-Text-01<\/a>, M1 uses a hybrid Mixture-of-Experts (MoE) architecture and a new \u201clightning attention\u201d mechanism. The model has a total capacity of 456 billion parameters, with 45.9 billion active per token, and supports context lengths of up to 1 million tokens.<\/p>\n

M1 distinguishes itself through its efficient use of compute and support for long-context reasoning. Its lightning attention mechanism reduces test-time computation, requiring only 25% of the FLOPs<\/a> used by DeepSeek R1<\/a> for sequences of 100K tokens. The model was trained using large-scale reinforcement learning across a range of domains, including mathematical problem-solving and software engineering environments.<\/p>\n

Two versions of the model are available. The models are evaluated using a custom RL scaling approach. Notably, MiniMax introduces CISPO, a novel RL algorithm that clips importance sampling weights rather than token updates\u2014reportedly improving stability and performance over traditional variants.<\/p>\n

Across benchmarks, MiniMax-M1-80K consistently ranks at or near the top among open-weight models, with strong results in:<\/p>\n