Elon Musk announces grok 4.8 training completion this week with 2.5 trillion parameters

Elon Musk, CEO of xAI, announced on the X platform that Grok 4.8 will complete its training this week. The model will feature 2.5 trillion parameters and will be trained using a new C++ software stack. Musk also outlined the next phase as reinforcement learning. Musk previously announced that Grok 4.7 would be released on September 12, but delayed the launch, citing issues with the reinforcement learning phase. He explained that the penalty for response length was too severe, causing the model to abandon difficult tasks and fail to review its work thoroughly. This was identified as a critical obstacle requiring resolution. The delay highlights the challenges in refining AI models, with Musk emphasizing the need for adjustments in the training process to ensure effective learning and performance.