MiniMax claims new M1 model needs half the compute of DeepSeek-R1

MiniMax claims new M1 model needs half the compute of DeepSeek-R1