Elon Musk said on social media that Grok 5 may surpass every AI product currently on the market, while outlining how he sees the capability progression of several upcoming Grok versions.
Musk outlines the next Grok versions
In his post, Musk wrote: "Grok 4.7 should be roughly equivalent to Opus 5.0, not 5.1. It is better in some respects and weaker in others. We need to improve multimodal performance. Grok 4.8 will bring a noticeable improvement. Grok 4.9 may reach Astra/Fable level. Grok 5 may surpass all current products. We shall see."
That places Grok 4.7, in Musk’s view, near the Opus 5.0 level rather than 5.1. He also said Grok 4.8 should deliver a clear step up, while Grok 4.9 could reach the Astra/Fable tier. His most aggressive claim was reserved for Grok 5, which he said may move ahead of all existing products.
Grok 4.8 to use a new C++ stack
Before this latest post, Musk had already previewed Grok 4.8. He said the model has 2.5 trillion parameters and is being trained on SpaceXAI’s new C++ software stack. Pretraining is expected to finish this week, after which reinforcement learning will begin.
Musk has repeatedly talked up that C++ stack before. In late June, he said Grok’s training and inference software would be substantially rewritten in C/C++ over roughly three months, with many intermediate layers removed and dedicated optimization for Nvidia GB300. He said that effort was expected to produce a "huge improvement" after about three months.
Grok 4.7 delayed over RL issues
The new comments also come after a delay to Grok 4.7 tied to RL issues. Musk had previously said the release would come around Sept. 12, but as launch approached, he acknowledged that the model was giving up too early on difficult problems and was not checking answers rigorously enough.
So while Grok 4.7 has not yet shipped on the earlier timeline, Musk is already sketching out expectations for Grok 4.8, Grok 4.9 and Grok 5.