MiniMax-M1, first open-weight, large-scale hybrid-attention reasoning model | Dark Hacker News