New MiniMax M2.7 proprietary AI model is ‘self-evolving’ and can perform 30-50% of reinforcement learning research workflow
