Grok 4.7

Name: Grok 4.7
Author: xAI

xAI · 2026-09-21

Grok 4.7 is SpaceXAI's September 2026 model for coding, agentic tasks, and knowledge work. It supports low, medium, high, and xhigh reasoning effort, with high as the default. Grok 4.7 Fast serves the same model on faster infrastructure. Its published knowledge cutoff is May 2026, without day precision.

Benchmark scores

Benchmark	Score
AA-3D v1.0 Elo	995
AA-AnalystAgent (pass@1)	67.0
AA-AnalystAgent (pass@5)	72.5
AA-AnalystAgent (pass^5)	55.0
AA-Briefcase v1.1 Elo	1638
AA-LCR v1.1	80.0
AA-Omniscience Index	32.0
AA-Omniscience v1.1 Accuracy	45.7
AA-Omniscience v1.1 Hallucination Rate	20.0
AA-Omniscience v1.1 Index	31.2
AdvancedIF	80.2
AdvancedIF (rubric mean score)	94.3
Aikido Known-CVE Detection (32 CVEs) single-run average	70.8
AIME 2025	100.0
APEX (Mercor)	16.3
APEX-Accounting	46.7
APEX-Accounting Pass@1	4.2
APEX-Agents 1.1	68.7
APEX-Agents 1.1 Corporate Law (68-task image-free slice)	75.6
APEX-Agents 1.1 Pass@1	54.6
APEX-SWE	53.4
APEX-v1-extended	65.4
ARC-AGI-1 Semi-Private	90.2
ARC-AGI-2	61.4
ARC-AGI-3 (Semi-Private)	10.0
Arena Alignment — Deceptive Completion	7.3
Arena Alignment — False Attribution	3.0
Arena Alignment — Unauthorized Action	1.5
Arena Alignment Index	82.7
Arena Code WebDev — Brand & Marketing	1645