Grok 4.7 Review

Grok 4.7 arrived on September 21, 2026 as xAI's strongest model for programming, agentic work and professional knowledge tasks. It is built on a larger base model than Grok 4.6 and sold at exactly the same price, $2 in and $6 out per million tokens, with a 500,000 token context and four levels of reasoning effort. A fast variant doubles output speed for double the price, but only inside Cursor and xAI's own Grok Build agent. The trap is in the pricing notes: a request above 200,000 tokens is billed in full at the higher rate, not just for the overflow.

Against its predecessor the gains are real. On xAI's own table Terminal-Bench 4.0 nearly doubles, from 20.3 to 37.6 percent, CursorBench 4.0 rises from 40.4 to 46.3, and it leads every listed rival on Harvey's legal agent benchmark and on an electrical engineering test. Against the leaders it is a step behind. Claude Fable 5.1 scores 57.9 on the same Terminal-Bench and 51.8 on CursorBench, and Anthropic's later Sonnet 5.5 reaches 70.6 on Terminal-Bench. xAI's comparisons also mix effort levels, Grok at xhigh against its predecessor at high, and an independent Proximal Labs run on its own harness measured a much smaller gain over Grok 4.6, 29.0 percent against 25.3.

What it offers is price. Output at $6 per million tokens is a fraction of what the frontier charges, which makes it a capable workhorse for long, repetitive agent runs, and it is easy to reach through the API, Cursor and Grok Build. It is not the model for the hardest coding or reasoning, and the long context bill needs watching, so it is a solid B.