Grok 4.7 Needs a Few More Days to Cook, Elon Says

Elon Musk said on September 11, 2026, that Grok 4.7 needs a few more days to cook, arguing that reinforcement-learning choices may have over-penalized response length so the model still gives up too early on hard tasks it can actually solve and is not yet rigorous enough about checking its own work.

“Grok 4.7 needs a few more days to cook,” Musk wrote in a reply on X. “We might have penalized response length too much (or something) in RL, as it still gives up on hard tasks (that it can do!) too early and isn’t yet sufficiently rigorous in checking its work.”

That is a concrete product-timing note from Musk rather than a vague tease: the next Grok flagship is close enough to talk about in days, not quarters, but the company is still tuning how long the model stays on a problem and how carefully it reviews its answers. Musk framed the remaining gap as an RL training tradeoff, not a missing feature list or a hardware bottleneck.

The update lands in a busy week for the wider Grok stack, including sales connectors for Salesforce, HubSpot, and Gong, the upcoming Grok Bot Galaxy sessions in San Francisco, and the recent inline drafting and usage upgrades. Those agent and workflow drops keep shipping while the foundation model Musk named as 4.7 gets a few more days of cooking before it is ready to replace whatever build users are on today.