On September 21, 2026 xAI released Grok 4.7, its third flagship in about ten weeks — after Grok 4.5 in July and Grok 4.6 in August. The change this time is under the hood: a new, larger base model and a longer reinforcement-learning run weighted toward tasks that take many hours, served at the same price and speed as Grok 4.6.
What changed
| Grok 4.6 | Grok 4.7 | |
|---|---|---|
| Price (in / out per 1M) | $2 / $6 | $2 / $6 |
| Context window | 500K | 500K |
| Terminal-Bench 4.0 (xAI table) | 20.3% | 37.6% |
| EEBench | 53.0% | 64.0% |
| CursorBench 4.0 | 40.4% | 46.3% |
| DeepSWE v1.1 | 65.2% | 71.0% |
Prompts above 200K tokens bill $4 / $1 / $12. A Grok 4.7 Fast variant — the same model at twice the output speed for 2x the token rates — runs only inside Cursor and Grok Build, where Grok 4.7 is the default model. There are four reasoning levels (low, medium, high and xhigh), and most headline numbers are at xhigh.
Why it matters
- Price-performance is the pitch. On CursorBench 4.0, xAI plots Grok 4.7 on the frontier of cost per task against Claude Fable 5.1, Opus 5, GPT-5.6 Sol and Sonnet 5. It costs a fifth of Claude Fable 5.1’s list price.
- Still not first. On xAI’s own table Fable 5.1 leads Terminal-Bench 4.0 (57.9% vs 37.6%) and CursorBench 4.0 (51.8% vs 46.3%), while Grok 4.7 matches GPT-5.6 Sol on terminal work and beats it on CursorBench, EEBench and the Harvey legal benchmark.
- A steady cadence at a flat price. Grok 4.5 (July 8), Grok 4.6 (August 12) and Grok 4.7 (September 21) arrived about every five to six weeks, each at $2 / $6 — so the capability jump comes without a price rise.
Caveats
- Every number is xAI-reported, and each lab runs rivals under its own harness, so independent tests matter.
- The 500K context window is half the 1M that Claude 5.5, GPT-6 and Gemini now offer.
- xhigh effort uses more reasoning tokens, so real cost per task can exceed the rate card.
For the wider model landscape, see the 2026 LLM guide.