Grok 4.7 is cheaper than Claude, but lags on performance

xAI, Elon Musk's AI company, launched a new version of its chatbot model Grok on September 21, 2026, called Grok 4.7. It is meant to be especially good at programming and knowledge-heavy work, that is, tasks where you ask the AI to analyze text or write code.

What stands out is the price. Companies that use Grok through a so-called API (a technical access point developers use to connect AI to their own programs) pay $2 per million words into the model and $6 per million words out. That is significantly cheaper than rival Claude from Anthropic, where according to several sources the price is considerably higher for a comparable task.

But cheaper is not the same as better. On an overall measure of artificial intelligence, the Artificial Analysis Intelligence Index, Grok 4.7 scores 46 points versus 53 points for Claude's newest model. That is a gap of 7 points, corresponding to a score about 13 percent lower.

The picture is mixed, however, when you look at individual tasks. On a test of programming skills, CursorBench, Grok 4.7 is roughly on par with Claude, while Grok performs significantly worse on a test of how well an AI can solve longer tasks in a computer terminal.

For ordinary users and companies, this in practice means Grok 4.7 can be a sensible choice if you need to run many large AI tasks cheaply, for example automated systems that use enormous amounts of text. If, on the other hand, you need the AI for tasks where precision and quality matter most, several analyses suggest that more expensive models such as Claude still deliver the better result.

Grok 4.7 is already available through the Grok app, the programming tool Cursor and xAI's own development tool Grok Build.

What this means for you

If you or your company need to run large volumes of AI tasks, Grok 4.7 can save you a lot of money over time. If, however, you need the most precise answers for important tasks, it is still worth paying more for Claude. You can already try Grok 4.7 in the Grok app or in Cursor if you code yourself. If you are a programmer, feel free to test it for code, but be aware that it performs worse on longer tasks in a terminal.