[Musk: Grok 4.8 Pre-training Completed, Limited Rollout to Begin Later This Week] Elon Musk posted an update on Grok 4.8: pre-training is complete, and it's currently in the reinforcement learning (RL) phase, with evaluation scores starting to rise. xAI plans to roll it out to a limited number of bot users (soft rollout) later this week. Musk stated that the team has abandoned JAX/SGLang and rewrote the entire reinforcement learning and inference system from scratch in C/C++, which is one of the reasons for the delays; supervised fine-tuning (SFT) data has also undergone extensive cleaning. He called Grok 4.8 "completely different" from 4.7 and "not yet outstanding," but the underlying foundation is in place.