xAI CEO Elon Musk announced that Grok 4.8, a model with 2.5 trillion parameters trained using a new C++ software stack, will complete training and initiate reinforcement learning this week. Meanwhile, the release of Grok 4.7, originally scheduled for September 12, may be delayed due to overly strict penalties for response length in reinforcement learning, causing the model to abandon complex tasks prematurely.