Musk: Grok 4.7 Coming in 10 Days, Set to Outperform All AI Models

Grok 4.7 will be publicly available in ten days and will outperform all existing AI models, according to Elon Musk.

This launch comes after Grok 4.6, which SpaceXAI released on August 12. The timing also coincides with OpenAI's announcement of its upcoming Astra model.

Grok 4.7 comes out in 10 days https://t.co/ZSXmzVFqB1

— Elon Musk (@elonmusk) September 2, 2026

SpaceX Data Plays Key Role in Grok 4.7

Musk has made his case through a series of posts. In mid-August, he stated that initial training was complete and that SpaceX data was being integrated via supplemental training.

Earlier, Musk outlined the model's architecture. Grok 4.7 operates on a 2.1 trillion parameter base, an increase from the 1.5 trillion parameters of Grok 4.6. According to him, the larger model is marginally slower but uses tokens more efficiently.

He also identified the competitor he anticipates lagging behind.

“Grok 4.7 will exceed all current models. That said, Anthropic is a great company and will probably release improved models soon. However, the SpaceX training corpus is so awesome & unique that I would be shocked if any model is better at real-world engineering than 4.7,” the post read.

The release schedule has become more frequent. SpaceXAI launched Grok 4.5 publicly in July and released Grok 4.6 on August 12.

Grok 4.6 Benchmarks Establish Standard

Grok 4.6 was based on the earlier Grok 4.5. According to the company, the model has a “particular focus on long-running agents and more ambitious interactive and visual work.”

Figures released show the model achieved a score of 61 on the AA Intelligence Index, tying with GPT-5.6 Sol Max and trailing Claude Fable 5 Max at 62. Grok 4.5 had scored 56.

Introducing Grok 4.6.

It delivers frontier intelligence and is a significant improvement over Grok 4.5 at the same price. pic.twitter.com/RtTbpXcb3a

— SpaceXAI (@SpaceXAI) August 12, 2026

Grok 4.6 topped GDPVal-AA v2 with a score of 1753. However, it achieved only 26% on Terminal-Bench v3.0, significantly lower than GPT-5.6 Sol Max's 34.6%.

A similar trend was seen with Grok 4.5. It led Artificial Analysis's AutomationBench-AA at 51.4% with a cost of $0.34 per task. But it recorded 0.63 guardrail violations per task, exceeding Claude Opus 4.8's 0.55.

Musk has now set a specific release date. The reach of his engineering claim will depend on whether SpaceXAI releases independent evaluations.