[Musk: Grok 4.8 Enters Reinforcement Learning Phase, Limited Rollout This Week] According to Jinse Finance, on November 10th, Musk responded to netizens on social media regarding the release date of Grok 4.8, stating: "We've abandoned JAX and SGLang and rewrote the entire reinforcement learning (RL) and inference technology stack from scratch in C/C++, which took some extra time. Training (i.e., pre-training) is complete. We did a lot of cleaning of the supervised fine-tuning (SFT) data to basically achieve the expected results. Currently, Grok 4.8 has entered the reinforcement learning phase, and various evaluation metrics are starting to improve. We plan to conduct a limited rollout to Bot users later this week to see how it performs. Grok 4.8 will be significantly different from 4.7. It's not amazing yet, but this time we've laid the right foundation, so I expect it to eventually develop into a truly unique product."