SpaceXAI released Grok 4.7 on Sept. 21 through its Grok Build coding agent, Cursor and the Grok API, with base prices matching Grok 4.6 at $2 per million input tokens and $6 per million output tokens. The company says the AI model works longer on difficult tasks, checks its work more carefully and handles long context better.

Grok Build with Grok 4.7 scored 56 on the Artificial Analysis Coding Agent Index, up from 47 with Grok 4.6 at the same reasoning setting. But unchanged token prices did not mean unchanged usage: on its separate Intelligence Index, Artificial Analysis measured about 81,000 output tokens per task for Grok 4.7 at xhigh reasoning effort, versus 36,000 for Grok 4.6 at high. Its example due diligence decks cost about $8 to produce with Grok 4.7, compared with $4.40 using Grok 4.6, both at xhigh.

Other tests gave a mixed picture. Cognition found Grok 4.7 slightly below its predecessor on FrontierCode 1.1, while Vals AI revised its Vals Index ranking to No. 10 from No. 24 after a software development kit update improved performance. Theo criticized its costs and frontend capabilities, saying, “Grok 4.7 is a disappointing release.” Elon Musk offered a different assessment: “Grok 4.7 is performing fairly well for a smallish model.”