Grok 4.7

What is Grok 4.7?

Grok 4.7 is a closed reasoning model from xAI, which now signs its docs as SpaceXAI. It uses a new, larger base model than Grok 4.6 and a longer reinforcement learning run on tasks that take hours, and xAI says it checks its own work more carefully and handles long context better. It reads text and images, writes text, works across a 500K-token context with four effort levels, and its knowledge stops at May 2026.

It is xAI's current flagship: the model its docs recommend for code and chat, the default in Grok Build, and available in Cursor and through OpenRouter, Vercel and Cloudflare. Grok 4.6 is still served at the same price.

What changed from Grok 4.6?

A bigger base model and longer training at the same price: xAI says it works longer on hard tasks and checks its own work more carefully.

Base model

A new, larger base model, trained longer with reinforcement learning on tasks that take many hours.

Coding scores

CursorBench 4.0 rises from 40.4% to 46.3%, and Terminal-Bench 4.0 from 20.3% to 37.6%, with 4.7 at xhigh effort and 4.6 at high.

Office work scores

GDPval rises from 1605 to 1695 Elo, and AA Briefcase v1.1 from 1546 to 1657.

Price and speed

Unchanged at $2 / $6 per million tokens, and xAI says it runs at the same speed.

Who is Grok 4.7 best for?

Grok 4.7 is best for coding agents on a budget, legal and electrical engineering work, and teams already using Cursor or Grok Build.

• Coding agents on a budget

  In xAI's table it beats GPT-5.6 Sol on CursorBench 4.0, 46.3% to 41.7%, at half the input price xAI lists for Sol.

• Legal work

  It leads xAI's table on the Harvey Legal Agent Benchmark at 19.6%, against 15.8% for Grok 4.6 and 6.7% for Fable 5.1.

• Electrical engineering

  It tops xAI's EEBench column at 64.0%, ahead of Fable 5.1 at 56.4% and GPT-5.6 Sol at 39.4%.

• Cursor and Grok Build users

  It is Grok Build's default model and on every Cursor plan, where a Fast variant runs at twice the token price.

What's included with Grok 4.7?

Four effort levels, function calling and structured outputs, image input, xAI's built-in tools, reasoning kept across turns, a US-only endpoint and priority processing.

Four effort levels

Reasoning effort runs low, medium, high or xhigh, with high as the default.

Function calling and structured outputs

It calls your own tools and returns answers in a fixed format, over the Responses or Chat Completions API.

Image input

It reads JPG and PNG images up to 20MiB each, with no limit on how many you send.

Built-in xAI tools

Web search and code execution run on xAI's servers for $5 per 1,000 calls, and X search for $5 per 1,000 posts fetched, on top of tokens.

Reasoning kept across turns

On the Responses API it always returns its encrypted reasoning, so a multi-turn agent keeps it by passing it back.

US regional endpoint

A US-only endpoint keeps inference in the United States for 10% more; only Grok 4.7 and Grok 4.6 run there.

Priority processing

Requests can jump the queue for lower latency at twice the standard token rates.

How much does Grok 4.7 cost?

Grok 4.7 costs $2 per million input tokens and $6 per million output tokens on the xAI API, rising to $4 and $12 once a prompt reaches 200K tokens.

Prices are in US dollars. Grok 4.7 runs on the xAI API as grok-4.7 and does not take Batch API requests. Priority processing costs twice the standard rates, and server-side tools cost extra, such as $5 per 1,000 web searches or code runs. Grok 4.7 Fast, the same model on faster hardware, is sold only in Cursor and Grok Build at $4 / $12 per million tokens, billed through those plans. xAI also offers Grok 4.7 to try free in Grok Build. Full pricing on docs.x.ai. We checked these prices on 29 Sep 2026.

How do readers rate Grok 4.7?

No one has rated Grok 4.7 yet. Be the first.

Grok 4.7 pros and cons

It scores well above Grok 4.6 at the same token price, but it uses far more tokens per task and still trails Fable 5.1 and Opus 5.5 on hard coding.

What are the best Grok 4.7 alternatives?

Four tools on Recaply do Grok 4.7's job. Ring-2.6-1T is open source, so you can run it on your own server.

Claude Sonnet 5Free

Choose Claude Sonnet 5 for long agent loops and long prompts: cache reads cost $0.20 per million tokens against $0.50 here, and its 1M-token context bills at a flat $2 / $10.

Claude Opus 5.5Paid

Choose it when coding quality matters more than price: Anthropic reports 57.8% on CursorBench 4.0 against 46.3% for Grok 4.7 in xAI's table, at $4 / $20 per million tokens.

Grok 4.3Paid

Choose Grok 4.3 for high-volume or batch work: it costs $1.25 / $2.50 per million tokens, reads a 1M-token context and takes Batch API jobs at 20% off.

Ring-2.6-1TOpen Source

Choose it when you want open weights with two reasoning effort levels for agent work, not xAI's hosted model.

Questions about Grok 4.7