Grok 4.8 model to finish training this week, Musk says
xAI's new 2.5T parameter model trained on C++ software stack enters reinforcement learning phase.
xAI's Grok 4.8 model will complete training this week and move into reinforcement learning, according to Elon Musk. The 2.5 trillion parameter model was built using xAI's new C++ software stack, a departure from the Python-based tools traditionally used in large language model development.
Reinforcement learning is a final training phase in which an AI model learns to improve its responses based on feedback, typically human evaluation of answer quality. The timeline puts the model's readiness for deployment sometime in the near term, though no public release date has been announced.
Grok, xAI's conversational AI, has been available to X Premium+ subscribers. Details on how Grok 4.8 will be deployed or what new capabilities it may offer remain unclear.
"@techdevnotes Grok 4.8, which is a 2.5T model trained with our new C++ software stack, will finish training this week and start RL"— X · @elonmusk