Overview
Grok 4.7 is the flagship model xAI released on September 21, 2026, aimed at coding, agentic tasks and knowledge work. It carries a 500,000-token context window with a knowledge cutoff of May 2026, takes text and image input, and returns text with no published output limit.
Pricing has two bands. Below 200,000 prompt tokens it is 2 dollars input, 0.50 dollars cached input and 6 dollars output per million tokens; above that it moves to 4, 1 and 12 dollars. Reasoning effort can be set to low, medium, high by default, or xhigh. On the tooling side it supports function calling, web search, X search, code execution and prompt caching.
In third-party testing the Artificial Analysis Intelligence Index scores it 46, two points above Grok 4.6. AA-Briefcase came in at 1657 Elo, up 111 points from the previous generation, and the coding-agent score rose to 56. Vendor figures include DeepSWE 1.1 at 71.08, CursorBench 4.0 at 46.33, Terminal-Bench 4.0 at 38.06, GDPval-AA v2 at 1695, HealthBench Professional at 56.72 and EEBench at 64. On the Artificial Analysis cost-per-task measure it lands at about 50 percent of Claude Opus 5, which Elon Musk cited in saying xAI now ranks third in agentic coding.
There is also a Grok 4.7 Fast variant: the same model on faster serving infrastructure, billed at twice the standard token rates, available only through Cursor and Grok Build and not on the public xAI API. It runs on the xAI API, as the default model in Grok Build, across all Cursor plans, and through the OpenRouter, Vercel and Cloudflare gateways. A US regional endpoint carries a 10 percent premium. To make cache hits reliable, the official guidance is to set a prompt_cache_key explicitly.
Key Features
- 500K-token context: Knowledge cutoff of May 2026, enough room to hold a large repository or a long multi-turn agent session.
- Coding and knowledge-work focus: Engineered around longer-running work with stronger self-checking and context management.
- Four reasoning levels: low, medium, high by default, or xhigh, letting cost and latency track task difficulty.
- Full tool surface: Function calling, web search, X search, code execution and prompt caching are built in.
- Encrypted reasoning passthrough: The Responses API always returns reasoning.encrypted_content, so multi-turn conversations keep the reasoning chain without extra configuration.
- Fast variant: Grok 4.7 Fast runs on faster infrastructure at twice the standard rates, limited to Cursor and Grok Build.
Use Cases
- Long-horizon coding agents that locate and fix defects across large repositories and self-check after cross-file edits
- Knowledge-work automation such as briefs, office documents and cross-industry analysis
- Legal and compliance triage covering the workflows the Harvey Legal Agent Benchmark measures
- Multi-turn sessions where an entire document and its reasoning history stay in context
Pros
- Coding-agent score rose to 56 and AA-Briefcase gained 111 points over the previous generation
- A 500K context paired with four reasoning levels lets one model cover different cost and latency targets
- Cost per task runs at about 50 percent of Claude Opus 5
- Available on the xAI API, in Grok Build, across Cursor plans and through three gateways
Pricing
Per million tokens: below 200,000 prompt tokens, 2 dollars input, 0.50 dollars cached input and 6 dollars output; above that, 4, 1 and 12 dollars. Grok 4.7 Fast bills at twice the standard rates and is served only in Cursor and Grok Build. The US regional endpoint carries a 10 percent premium. Check official listings for current rates.
Summary
Grok 4.7 is the xAI flagship for coding, agentic tasks and knowledge work, with a 500K-token context, text and image input, four reasoning levels and built-in function calling, web search, X search, code execution and prompt caching. It scores 46 on the Artificial Analysis Intelligence Index, 1657 Elo on AA-Briefcase, up 111 points, and 56 on the coding-agent measure, at roughly half the cost per task of Claude Opus 5. Pricing is 2 dollars input and 6 dollars output per million tokens below 200,000 prompt tokens, doubling above that.
Version History
- xAI releases Grok 4.7 for coding and knowledge work (2026-09-21): xAI released Grok 4.7, positioned as its flagship model for coding, agentic tasks and knowledge work, with a 500,000-token context window, a May 2026 knowledge cutoff, text and image input, text output and reasoning effort selectable across low, medium, high and xhigh. Pricing is 2 dollars input and 6 dollars output per million tokens below 200,000 prompt tokens, rising to 4 and 12 dollars above that. The Artificial Analysis Intelligence Index scores it 46, two points above Grok 4.6, AA-Briefcase reached 1657 Elo for a gain of 111 points, the coding-agent score rose to 56, and cost per task runs at about 50 percent of Claude Opus 5. A Grok 4.7 Fast variant of the same model bills at twice the standard rates and is available only in Cursor and Grok Build.