Grok 4.7

xAI's new model is here. The company announced Grok 4.7 on 21 September 2026. For coding and office work, the model takes over from Grok 4.6. It is built on a new base model, bigger than the one under Grok 4.6. After that it went through a longer round of reinforcement learning (training by reward), and that round puts the weight on hard tasks that run for hours.

The price hasn't moved: 2 dollars per million input tokens, 6 dollars per million output tokens. For requests of 200 thousand tokens and up, those figures climb to 4 and 12 dollars. On top of that, the higher rate applies to the whole request. The context window (the amount of text the model can read in one go) is 500 thousand tokens. You can use the model in the Grok API, in Cursor and in xAI's own Grok Build tool.

In xAI's tables the jump looks pretty big. Mind you, 4.7 was measured at its highest setting and 4.6 one setting below. The independent measurements are more modest. Artificial Analysis gives the model 46 points on its intelligence index; that is about two points above Grok 4.6. The real flaw is that it never stops talking. Across the index's tasks the model produced 240 million output tokens in total (the median is 81 million). The cost per task comes to about 3.24 dollars. The sticker price is cheap, but the bill isn't quite so cheap. Do your sums accordingly.