Elon Maps the Grok Roadmap Through Grok 5 After 4.8 Training

Elon Musk laid out xAI’s Grok model ladder on September 14, 2026, saying Grok 4.8 will finish training this week and move into reinforcement learning, ranking Grok 4.9 as roughly Astra/Fable class, and answering an AGI question by pointing past 4.8 to Grok 5 while adding that Grok 5 “maybe better than anything.” The timeline lands against a fresh safety-and-policy backdrop: Musk spent the weekend backing Anthropic CEO Dario Amodei’s call to pace frontier AI, while President Donald Trump rejected new AI “guardrails” on Monday.

“Grok 4.8, which is a 2.5T model trained with our new C++ software stack, will finish training this week and start RL,” Musk wrote on X in reply to a question about what else xAI has coming in September beyond Grok 4.7. Asked how close 4.8 would be to AGI, he answered simply, “That will be Grok 5.” In a longer follow-up, he said Grok 4.7 should be roughly on par with Opus 5.0 rather than 5.1, with multimodal performance still needing work, called 4.8 a noticeable improvement, put 4.9 in the Astra/Fable class, and left Grok 5 as “maybe better than anything. We shall see.”

Two days earlier, Amodei published “We Must Pace the Frontier”, arguing the industry should slow how fast it improves model capabilities so alignment and third-party checks can keep up. Anthropic said it would unilaterally give embedded third-party evaluators employee-like access. Musk quoted the announcement and wrote, “Dario is right.” The next day he added that Amodei was “right that there should be some oversight” and that “peer review of AI by competitors is the right way to start this off.”

President Trump pushed the other direction on Monday in a Truth Social post, writing that the only control or “guardrails” AI needs is “a STRONG AND SMART (High IQ!) PRESIDENT,” naming Amodei as someone “now pretending to be a ‘perfect little angel,’” and arguing the United States already has “tremendous CRIMINAL and REGULATORY power” over AI companies. He framed the stakes as geopolitical: “WHOEVER WINS AI, WINS!”

That is why today’s roadmap matters beyond a training update. Musk is publicly endorsing paced oversight while still sketching an aggressive public ladder: finish 4.8’s 2.5 trillion-parameter training and start RL this week, treat 4.9 as the next competitive rung, and reserve Grok 5 for the AGI-level claim. The sequence also sits on top of last week’s Grok 4.7 “needs a few more days to cook” note about reinforcement-learning choices that may have over-penalized response length.

None of the posts named a Grok 5 launch date or benchmark board, so the useful near-term watch is whether 4.8 completes training and enters RL on the timeline Musk described, whether 4.7’s multimodal fixes land before that handoff, and whether xAI restates the Grok 5 ambition (or any peer-review posture matching the oversight language Musk backed) once 4.8 is in RL.