Grok 4.7

Grok 4.7 shipped on September 21, 2026 as grok-4.7, and xAI now recommends it for everything outside its specialized models, code included. The price did not move: $2 per million input tokens, $0.50 cached, and $6 output below 200,000 prompt tokens, doubling to $4, $1, and $12 once a prompt reaches 200,000, with the higher rate applied to every token in that request. What changed is underneath. Grok 4.6 was a post-training upgrade on the 1.5 trillion parameter V9 foundation; 4.7 moves to a new, larger base model (reported at 2.1 trillion parameters, a figure xAI has not published itself) with a longer reinforcement learning run weighted toward tasks that take hours. The context window stays at 500,000 tokens, input is text and image, reasoning effort runs low, medium, high, and xhigh with high as the default, and the Batch API is not supported. On xAI's own table it posts 46.3 percent on CursorBench 4.0 against 40.4 for Grok 4.6, 71.0 percent on DeepSWE v1.1 at high effort, and 38.0 percent on Terminal-Bench 4.0, where Fable 5.1 at max effort reaches 57.9. Independent numbers are lower and more useful: the Artificial Analysis Intelligence Index scores it 46, twelve points behind the leader, and the Vals Index places it tenth at 60.22 percent. That is the honest read on Grok 4.7: not a frontier leader, but a frontier-adjacent model at a third of GPT-5.6 Sol's promotional rate and a fifth of Fable 5.1's. A fast variant runs at twice the output speed for twice the price. Available in Cursor, Grok Build, the Grok API, and the major routers and clouds.