Elon Musk's next Grok model is late, and this time he explained why. On September 11, 2026, the day his own timeline pointed to, Musk posted that Grok 4.7 needs a few more days to cook. He had written on September 1 that the model comes out in 10 days.
What went wrong
Musk's explanation was unusually technical. He said the team may have penalized response length too much during reinforcement learning, with the result that the model still gives up on hard tasks it is actually capable of solving, and that it is not yet rigorous enough about checking its own work. In other words, the remaining fixes are training adjustments, not missing features or hardware.
Musk has previously said Grok 4.7 is a 2.1 trillion parameter model trained in part on SpaceX engineering data. As of this weekend, xAI had not published a launch page, API model identifier, price, model card or benchmark table for it. The current flagship remains Grok 4.6, released on August 12.
A busy year for xAI
The delay comes amid structural change at the company. SpaceX acquired xAI in February 2026, and the combined entity was formally renamed SpaceXAI in August. On the product side, xAI recently launched Grok Bot for enterprises, with a free trial and persistent cloud agents, and Tesla's summer software update put Grok Voice into its cars.
Musk has also raised the stakes for what follows. In September he said he now believes xAI could reach artificial general intelligence with Grok 5, a milestone he previously considered unlikely, and that Grok 5 is targeted before the end of 2026.
Why it matters
Musk's shipping dates have slipped before, and a few days of delay is minor on its own. What stands out is the candor about reward design, a topic AI labs rarely discuss in public. If Grok 4.7 arrives with the promised fixes, it will be judged against GPT-6 Astra and Claude Fable 5.1, both released this month.
Source: TeslaNorth, DataStudios
Photo: Tesla North (source)
