xAI · current

Grok 4.7

Grok 4.7 is xAI’s frontier model for coding, agentic tasks, and knowledge work. The API id is grok-4.7. It shipped on 21 Sep 2026 for Cursor, Grok Build, the Grok API, and partner gateways. Consumer web, mobile, and Grok on X come later, per the model card.

Status: Current
AA score: 46 (xhigh effort)Artificial Analysis Intelligence Index, one independent score across many tests, fetched 22 Sep 2026
Price: $2 in, $6 out (under 200k)US dollars per million tokens (word pieces) in and out, unless noted
Released: 21 Sep 2026

Training

xAI describes a larger base than Grok 4.6, with a longer reinforcement-learning run on harder multi-hour tasks, stronger self-checking, and better long-context management. The model was also trained to work natively with the Grok Bot harness. The model card notes supplemental training on anonymized Cursor workflow data. Pretraining data cuts off in June 2026. Supplemental training uses data generated as late as August 2026. Parameter count and training FLOPs are not published on the announce or docs pages.

Inference

Reasoning effort can be set to low, medium, high (default), or xhigh. The model accepts text and images and returns text. Function calling and structured outputs are supported. Batch API is not. xAI also serves a faster variant at twice the output speed and twice the price. Docs list rate limits of 150 requests per second and 50 million tokens per minute, in us-east-1, us-west-2, and us-central-1.

Context

The documented context window is 500,000 tokens.

Methods

Grok 4.7 uses configurable reasoning effort for longer agent runs. The model docs list function calling and structured outputs. Reasoning effort can be set from low through xhigh.

Benchmarking

xAI publishes side-by-side scores on the announce page, including CursorBench 4.0, DeepSWE v1.1, EEBench, AA Briefcase v1.1, Terminal-Bench 4.0, the Harvey Legal Agent Benchmark, and HealthBench Professional. On that table, CursorBench 4.0 is 46.3% for Grok 4.7 versus 40.4% for Grok 4.6. DeepSWE v1.1 is 71.0% at high effort for Grok 4.7 versus 65.2% for Grok 4.6. The announce and model card are the sources for the figures above. The model card adds further coding, engineering, safety, and cyber results.

Hardware

xAI has not published training or serving silicon details for Grok 4.7 on the announce or model docs. Campus-scale hardware belongs on the separate hardware map when those facts are public.

Access and pricing

Under 200k prompt tokens, docs list $2.00 input, $0.50 cached input, and $6.00 output per million tokens. At or above 200k prompt tokens, those rates become $4.00, $1.00, and $12.00 for the whole request. The announce also describes a fast variant at twice the price for twice the output speed.