Musk says Grok 4.8 has finished pre-training, is now in RL, and gets a soft rollout later this week

Elon Musk said on X at 09:35 UTC on November 10 that Grok 4.8 has finished pre-training and is now in reinforcement learning, and that xAI will do a soft rollout to Bot users later this week. xAI has given no model id, price or exact date, and no grok-4.8 is on its API yet.

WHAT MUSK SAID

• xAI dropped JAX and SGLang and rewrote its whole RL and inference stack from scratch in C and C++, which he says took extra time.
• Pre-training is done, and the SFT data needed a lot of cleanup.
• Grok 4.8 is now in RL, and he says its evals are starting to climb.
• He says 4.8 will be nothing like 4.7 and is not yet incredible, but that the foundations are now built right, so he expects it to grow into something special.

THE CONTEXT

• Grok 4.7 shipped on September 21, 50 days ago. The gaps before it were 34 days from Grok 4.5 to 4.6 and 40 days from 4.6 to 4.7.
• On October 7 Musk said most Grok Bot requests would be handled by a fast version of Grok 4.8 once it comes out. On October 30 he said 4.8 had taken longer than expected.
• Polymarket's market on the next Grok (4.8 or later) arriving by November 30 traded at 63% at 10:51 UTC, little changed from 64% on November 7. The December 7 market traded at 80%.

WHAT WE DO NOT KNOW

Who counts as a Bot user, whether the soft rollout reaches xAI's API, the model id, the price and any benchmarks. At 10:51 UTC OpenRouter's newest xAI model is still grok-4.7. ArtificialWatch's Grok 4.8 hunt is armed and sends an alert the moment a grok-4.8 id is served.