Grok 4.8 finally has a rollout window. Elon Musk says xAI plans a soft rollout to Bot users later this week, with pre-training finished and the model now in reinforcement learning. It’s a plan for an initial rollout, rather than a date for a public API release. 1
I’d literally just posted asking whether we were giving up on 4.8, so the timing was pretty good. Grok 4.7 didn’t feel like the same jump we got from 4.6, and I’ve been hoping the next one would have more behind it. 2 3
Musk says the delay involved a substantial rewrite: xAI dropped JAX/SGLang and rebuilt its reinforcement-learning and inference stack in C/C++. He also describes cleanup of the supervised fine-tuning data before moving on to this stage of training.
We will do a soft rollout to Bot users later this week and see how it goes.
Elon Musk · 10 November 2026
He says the evaluations are starting to climb, but also calls the model “not yet incredible”. That makes this an interesting progress update, with work still going on. There aren’t public evaluation figures in the statement to show how large the improvement is.
I’m hopeful. A bigger jump would make Grok much more interesting to test again, especially after the wait. For now, the concrete thing to watch is whether the promised soft rollout reaches Bot users this week. We’ll get a much better read on the model once people can actually use that version.
Sources & references
- Elon Musk: Grok 4.8 training and soft-rollout update ↗
10 November 2026, 9:35 am. Musk describes completed pre-training, ongoing reinforcement learning and a planned soft rollout to Bot users. This is not a general API release date.
- Lumina’s earlier Grok 4.8 post ↗
10 November 2026. The post asking for an update shortly before Musk’s statement.
- Lumina: reaction to the Grok 4.8 update ↗
10 November 2026. Luke’s hopes for a bigger improvement over 4.7, distinct from measured evaluation results.
