Grok 4.6

Name: Grok 4.6
Author: xAI

xAI · 2026-08-12

Grok 4.6 is SpaceXAI's August 2026 model for coding, agentic tasks, and knowledge work, developed through additional training beyond Grok 4.5. It accepts text and image inputs, produces text, and supports low, medium, high, and xhigh reasoning effort with high as the default.

Benchmark scores

Benchmark	Score
AA-AnalystAgent (pass@1)	60.8
AA-AnalystAgent (pass@5)	76.3
AA-AnalystAgent (pass^5)	41.3
AA-Briefcase Elo	1545
AA-Briefcase v1.1 Elo	1541
AA-LCR v1.1	81.0
AA-Omniscience Index	30.5
AdvancedIF	80.9
AdvancedIF (rubric mean score)	95.5
Agent Red Teaming k=100 (CAIS)	47.1
AHK-Eval	97.2
AHK-Eval Algorithms	83.3
AHK-Eval Data Structures	100.0
AHK-Eval Date and Time	100.0
AHK-Eval Easy	100.0
AHK-Eval Hard	91.7
AHK-Eval Hidden Cases	97.2
AHK-Eval Mid	100.0
AHK-Eval Numbers	100.0
AHK-Eval Parse Success	100.0
AHK-Eval Regex	100.0
AHK-Eval Strings	100.0
AIME 2025	100.0
ALE-Bench	1822
Antidote: Everyday Edition	966
APEX (Mercor)	17.6
APEX-Agents (release-table score, metric unspecified)	57.5
APEX-Agents 1.1	77.3
APEX-Agents 1.1 Corporate Law (68-task image-free slice)	82.2
APEX-Agents 1.1 Pass@1	65.3