September 22, 2026
Grok Build Promises Everyday Work: Grok 4.7 Is Its Default
On September 22, Elon Musk called Grok 4.7 with Grok Build an everyday work tool. A day earlier, Artificial Analysis gave the pairing 56 points in the Coding Agent Index; Grok 4.6 had 47.
Elon Musk
@elonmusk
Grok 4.7 with our Build setup confidently handles daily work.
I’ve been building with Grok 4.7 extra high in Grok Build today for hours. It’s fantastic and my usage percent (Cursor Ultra) is barely moving. Getting so much done. While Build works on one thing, I discuss the next feature/change with my Grok Bot team and have Engineer Bot
· 5.3M views
Previously, Grok 4.6 xhigh scored 47 points in the Coding Agent Index. Grok 4.7 xhigh with Grok Build scored 56. On DeepSWE, the result rose from 65% to 73%; on Terminal-Bench, from 18% to 33%.
To achieve this improvement, Grok 4.7 in the September 21 benchmark spent about 81 thousand output tokens per task, compared with 38 thousand for Grok 4.6. In the API, one million input tokens cost $2, and one million output tokens cost $6.
Terminal launch. Grok Build works as an AI coding agent in the terminal. On macOS, Linux, and WSL, it is installed with `curl -fsSL https://x.ai/cli/install.sh | bash`, then `grok` is run in the project directory. On first launch, the agent opens a browser for authentication.
Grok 4.7 Fast is already available in Cursor and Grok Build, runs twice as fast, and costs twice as much.
Source
