Musk reposted a post from XFreeze stating that pre-training for Grok 4.8 is complete, supervised fine-tuning data has undergone large-scale cleaning, reinforcement learning is currently underway, and a phased rollout to Grok Bot users is planned for later this week. The post also noted that the team has rewritten the reinforcement learning and inference infrastructure in C/C++ to replace JAX and SGLang. In a separate post on X, Musk stated that the software used by Starlink to manage massive amounts of data between space and ground is also exceptionally well-suited for training superintelligence. The aforementioned progress and technical statements come from relevant posts.