‹ PublicAI Index
The LLM benchmark aggregator.
Grok 3
SpaceXAI
Strongest in Factual grounding (#23 of 104), weakest in Safety (#289 of 329). Above par in 13 of 24 scopes. Among the models it meets almost everywhere, it finishes behind O3 and Claude Opus 4.8 and ahead of Qwen3.5 Flash and GPT-4 Turbo.
§ 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.
Professional52.4−10#65/1941/2
Legal53.3−12.6#63/1731/1
Finance54−8.4#65/1621/1
Medical50.2−16.3#86/1491/1
Core abilities52.9−12.4#68/1991/4
General intelligence54.2−15.2#56/1781/3
Coding52.1−14.1#71/1791/6
Agentic coding52.2−14.7#67/1731/5
Knowledge49.3−13.2#121/1971/3
Academic knowledge48.4−15.5#85/1231/1
Human preference57.7−11.1#122/3541/1
Human preference57.7−11.1#122/3541/1
Reasoning44.9−20.5#143/1953/5
Science48.7−14.8#82/1221/1
Mathematics48−15.8#95/1521/2
Reasoning40.4−28.6#120/1472/3
Safety44.1−17.3#289/3293/3
Factual grounding60.5−10.2#23/1041/1
Safe-prompt compliance55.1−5.3#34/821/1
Fairness51.4−18.6#96/2922/2
Secure code40.8−25.4#211/2701/1
Harm refusal43.5−18#259/2922/2
Toxicity avoidance26−33#263/2681/1
Jailbreak resistance26−39#268/2681/1
§ 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 3, left for the other.
§ 3 · Sources
Where the numbers come from
9 publications, 24 figures. Every one links to the page it was read from.
LMArena Text 1411
ARC-AGI-2 0%
Aider polyglot 53.3%
Kagi LLM Benchmark 61.3%
SimpleBench 36.1%
Vals · LegalBench 82.59%Vals · CorpFin 59.71%Vals · TaxEval 75.88%Vals · MedQA 83.85%Vals · GPQA Diamond 74.24%Vals · MMLU Pro 79.95%Vals · AIME 58.75%
HELM Safety · HarmBench 45.3%HELM Safety · SimpleSafetyTests 96.8%HELM Safety · Anthropic Red Team 95.5%HELM Safety · BBQ 93.6%HELM Safety · XSTest 96.4%
Enkrypt · Jailbreak risk 68.2%Enkrypt · Harmful content risk 59.4%Enkrypt · CBRN risk 12%Enkrypt · Toxicity risk 38.3%Enkrypt · Bias risk 80.6%Enkrypt · Insecure code risk 52%
Vectara · Factual consistency 94.2%
Badge
[](https://publicai.io/model-index/m/grok-3)