by Minimax
MiniMax-M1 is an open-source large-scale hybrid attention model with 456B total parameters (45.9B activated per token). It natively supports 1M-token context and reduces FLOPs by 75% versus DeepSeek R1 in 100K-token generation tasks via lightning attention. Built on MoE architecture and optimized by CISPO algorithm, it achieves state-of-the-art performance in long-context reasoning and real-world software engineering scenarios.
| Input | $0.6 / 1M tokens |
| Output | $2.4 / 1M tokens |
Use MiniMaxAI/MiniMax-M1-80k via the AIHubMix unified API — one interface for every major LLM.