Grok 4.20 Thinking model using reinforcement learning to think longer about problems and solve them.