xAI Grok 1.5T enters reinforcement learning

xAI is moving at breakneck speed to train and align extremely large-scale models, but reinforcement learning on a 1.5T model is a massive computational hurdle that will test the limits of their GPU clusters.

• –**Compute Intensity:** Conducting RL on a 1.5-trillion parameter model requires an immense amount of high-bandwidth memory and computing power, meaning xAI is utilizing their massive infrastructure to its full capacity.
• –**Release Timeline:** Transitioning to RL suggests that the base model is fully cooked, pointing to a potential release within the next few months if safety and alignment tuning goes smoothly.
• –**Competitive Landscape:** A 1.5T parameter model would put Grok in direct competition with frontier models from OpenAI and Anthropic in terms of raw capacity and reasoning capabilities.