88Z AI, 12 November — Replying on X to a question about Grok 4.8's timing, Musk said pre-training is finished, the RL and inference stack was rewritten from scratch in C/C++, and the model is now in RL with evaluations just beginning.
At a glance
- Musk confirms Grok 4.8 pre-training is done; the model is now in RL with evaluations just starting.
- The team dropped JAX/SGLang and rewrote the RL and inference stack in C/C++, which added time.
- No release date or evaluation data was given, so readers should wait for official follow-ups.

Pre-training finished, now in RL
On 10 November 2026, Elon Musk replied on X to a public question about when to expect Grok 4.8. He said the team dropped JAX/SGLang and rewrote the entire RL and inference stack from scratch in C/C++, which took extra time. Pre-training is done, he said, and the SFT data needed a lot of cleanup to get mostly right. Grok 4.8 is now in RL and the evaluations are starting to — the post's public text ends there, with no release date and no evaluation results given.
How to read this
This is a progress update, not a launch announcement. The post mentions no availability date, no rollout scope and no benchmark figures, so it is not a basis for predicting when Grok 4.8 will reach users. Readers can watch for follow-up statements from Musk and xAI for evaluation results or release plans.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- @elonmuskSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.