‹ PublicAI Index
The LLM benchmark aggregator.
Grok 4.6
SpaceXAI
Strongest in Secure code (#2 of 270), weakest in Toxicity avoidance (#125 of 268). Above par in 30 of 33 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5.1 and Claude Opus 5.5 and ahead of MiMo V2.6 Pro and Gemini 3.1 Pro.
§ 1 · Profile
What it is good at
Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.
Safety58.2−3.2#17/3291/3
Secure code66−0.2#2/2701/1
Fairness70leads#7/2921/2
Jailbreak resistance56.6−8.4#91/2681/1
Harm refusal53.3−8.2#100/2921/2
Toxicity avoidance54.2−4.8#125/2681/1
Reasoning57.8−7.6#23/1955/5
Science63−0.5#3/1221/1
Reasoning61.1−7.9#18/1473/3
Mathematics51.5−12.3#77/1522/2
Coding56.8−9.4#24/1794/6
Agentic coding57−9.9#24/1734/5
Professional55.7−6.7#27/1942/2
Document design55.1−13.6#11/311/1
Biology research53.8−16.9#12/311/1
Legal59.1−6.8#17/1731/1
Cybersecurity52.5−11.4#29/471/1
Medical55.1−11.4#40/1491/1
Finance55.6−6.8#41/1621/1
Knowledge54.3−8.2#42/1971/3
Academic knowledge60.2−3.7#11/1231/1
Human preference62−6.8#58/3541/1
Human preference62−6.8#58/3541/1
Core abilities53.7−11.6#60/1992/4
Language57.4−9.2#16/621/1
Data analysis52.4−12.8#19/311/1
Instruction following56.9−16.9#20/621/1
General intelligence49−20.4#99/1781/3
Agents54.8−11.6#61/2983/6
Web research58.2−5.7#7/351/2
Knowledge work65.5−5.2#17/2082/2
Long-horizon work39.8−34.2#24/311/1
Computer use46.9−21.2#42/681/2
§ 2 · Head to head
What it beats, and what beats it
The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Grok 4.6, left for the other.
§ 3 · Sources
Where the numbers come from
10 publications, 44 figures. Every one links to the page it was read from.
LMArena Text 1456
GDPval-AA 1654
AA-Briefcase 1541
Terminal-Bench 19.7% (Grok Build)
ARC-AGI-2 67.1%
LiveBench 63.5LiveBench · Agentic 18.7LiveBench · Reasoning 91.2LiveBench · Coding 76.7LiveBench · Browser Use 49.4LiveBench · Data Analysis 75.3LiveBench · Research 88.8LiveBench · Design 76.3
LiveBench 1.0 78LiveBench 1.0 · Reasoning 90.5LiveBench 1.0 · Coding 76.8LiveBench 1.0 · Agentic Coding 57LiveBench 1.0 · Mathematics 92.6LiveBench 1.0 · Data Analysis 73.9LiveBench 1.0 · Language 83.7LiveBench 1.0 · Instruction Following 71.9
SimpleBench 75.9%
Vals · Legal Research Bench 48.08%Vals · LegalBench 86.31%Vals · Finance Agent 53.68%Vals · CorpFin 66.16%Vals · TaxEval 71.1%Vals · MortgageTax 64.19%Vals · MedCode 44.71%Vals · MedScribe 86.53%Vals · BioMysteryBench 72.22%Vals · CyberBench 67.68%Vals · SWE-bench Verified 95.6% (Mini-SWE-agent)Vals · Vibe Code Bench 76.24% (OpenHands)Vals · Code Migration 44.57%Vals · GPQA Diamond 94.7%Vals · MMLU Pro 89.4%Vals · ProofBench 51%
Enkrypt · Jailbreak risk 7.3%Enkrypt · Harmful content risk 0.6%Enkrypt · CBRN risk 24%Enkrypt · Toxicity risk 3.3%Enkrypt · Bias risk 35.1%Enkrypt · Insecure code risk 0.4%
Badge
[](https://publicai.io/model-index/m/grok-4-6)