Elon Musk: Grok 4.8 Finished Pretraining and Starts a Soft Rollout This Week
We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time. Training (aka pre-training) is done. Lot of cleanup was needed on the SFT data to get that mostly right. Now, 4.8 is in RL and the evals are starting to climb. We will do a soft rollout to Bot users later this week and see how it goes. Grok 4.8 will bear no resemblance to 4.7 (thankfully). It is not yet incredible, but, this time, the foundations are built right, so I expect it to evolve to be something special.
Context
It is the first concrete timeline for xAI's next Grok model, and it comes with a full rewrite of the training and serving stack behind it.
Grok is the AI model family built by Elon Musk's AI team, which posts now describe as part of SpaceX. Grok 4.7 is the current version, and users had been asking Musk when 4.8 would arrive. Musk answered one of those questions directly on X.
Deep dive
Asked on X when to expect Grok 4.8, Elon Musk gave a progress update on the model.
Musk wrote that the team dropped JAX and SGLang and rewrote the entire reinforcement learning and inference stack from scratch in C and C++, which he said took extra time. JAX is a machine learning framework from Google and SGLang is an open source serving engine for large language models, so the change means the team now runs its own code for both training by reinforcement learning and serving the model.
He said pretraining is done and that a lot of cleanup was needed on the supervised fine tuning data. According to Musk, Grok 4.8 is now in reinforcement learning and its evaluation scores are starting to climb. He said there will be a soft rollout to Bot users later this week.
Musk added that Grok 4.8 will bear no resemblance to 4.7, which he followed with the word thankfully. He described it as not yet incredible but said the foundations are built right this time, so he expects it to become something special. The account X Freeze summarized the update in a widely shared post, and Musk reposted it.
Musk did not share benchmark numbers or a date for a wider release.
What to expect next
The soft rollout to Bot users later this week is the first test. Watch for xAI benchmark results, early user comparisons against Claude Fable 5.2 and the latest OpenAI and Google models, and whether 4.8 reaches the Grok app and API soon after. Musk's own words set expectations low for the first version, so the bigger signal may be how fast the new C and C++ stack lets the team ship updates in the weeks after launch.