Recently, AI startup MiniMax launched its first inference model M1. According to benchmark evaluation results, M1 has excellent performance, surpassing domestic closed source models, approaching the level of the most advanced models overseas, and even surpassing domestic companies such as DeepSeek, Alibaba, ByteDance, as well as the latest and strongest open closed source models from foreign companies such as OpenAI, Google, and Anthropic in some tasks.
In comparison with DeepSeek R1, M1 exhibits a significant cost advantage in computing power. When generating 64K tokens, M1 consumes less than 50% of DeepSeek R1's computing power; When generating 100K tokens, this proportion drops to 25%.
MiniMax stated that the entire reinforcement learning process of M1 only used 512 NVIDIA H800 GPUs for three weeks of training, with a rental cost of $537400 (approximately 3.8 million RMB). This cost control effect is' one order of magnitude less than initially expected '.
