SpaceX logo (public domain) via Wikimedia Commons
Elon Musk says the 2.1 trillion parameter model needs more reinforcement-learning tuning as competitors Meta and OpenAI prepare September launches
Elon Musk’s xAI has pushed back the release of Grok 4.7, its next-generation AI model, citing the need for additional reinforcement-learning tuning. The delay extends a pattern of slipping timelines that stretches back to late July, and it arrives at a particularly inconvenient moment: both Meta and OpenAI are expected to drop competing model updates this month.
Musk indicated on September 11 that Grok 4.7 needs “a few more days” of RL tuning to resolve issues with response length penalties and task management.
A timeline that keeps sliding #
On July 24, Musk suggested Grok 4.7 would land “in 4 weeks.” Four days later, on July 28, he revised that to “a few weeks” after Grok 4.6’s release. When Grok 4.6 actually shipped on August 12, Musk pegged the successor at “3 to 4 weeks” out.
Every single one of those windows has now closed. The September 11 update, which originally appeared to be the target date, instead became the date of yet another delay announcement.
What we know about Grok 4.7 #
Almost everything known about the upcoming model comes from Musk’s social media posts rather than official xAI documentation. No technical paper, model card, or benchmark suite has been published.
What Musk has shared: Grok 4.7 is expected to contain 2.1 trillion parameters, a significant jump from Grok 4.6’s 1.5 trillion. That’s a roughly 40% increase in raw model size. Musk has said the new version should outperform its predecessor “in most areas” but will be “slightly slower to serve,” which tracks with the parameter increase.
Grok 4.6, the current flagship, launched on August 12 and has been available to users on the X platform.
The competitive pressure is real #
The timing of this delay matters because Meta and OpenAI are both expected to release updated models in September, which means xAI risks falling behind in the news cycle and, more importantly, in capability comparisons.
There’s also the credibility question. When a CEO publicly sets a timeline four separate times and misses all four, the fifth estimate gets discounted. “A few more days” could mean anything from 72 hours to another month.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our