‹ PublicAI Index
The LLM benchmark aggregator.
GPT-6 Sol
OpenAI
Strongest in Cybersecurity (#2 of 47), weakest in Toxicity avoidance (#222 of 268). Above par in 26 of 28 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5 and Claude Fable 5.1 and ahead of Gemini 3.8 Flash and Claude Opus 4.7.
§ 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.
Reasoning60.5−4.9#12/1954/5
Mathematics60.2−3.6#12/1522/2
Reasoning62−7#15/1472/3
Coding58.3−7.9#18/1793/6
Agentic coding58.7−8.2#17/1733/5
Agents58.3−8.1#26/2982/6
Knowledge work63.3−7.4#32/2082/2
Knowledge56.3−6.2#30/1971/3
Document parsing56.7−6.6#28/831/1
Professional54.5−7.9#36/1941/2
Cybersecurity61.8−2.1#2/471/1
IT operations62.1−11.7#6/331/1
Biology research57.6−13.1#9/311/1
Medical54.4−12.1#50/1491/1
Finance50.9−11.5#93/1621/1
Legal48.3−17.6#103/1731/1
Human preference62.1−6.7#56/3541/1
Human preference62.1−6.7#56/3541/1
Core abilities53.8−11.5#59/1991/4
Language59.5−7.1#13/621/1
Instruction following51.1−22.7#32/621/1
Safety52.4−9#117/3292/3
Factual grounding58.7−12#29/1041/1
Fairness54−16#77/2921/2
Secure code57.3−8.9#92/2701/1
Harm refusal50.5−11#145/2921/2
Jailbreak resistance50.7−14.3#173/2681/1
Toxicity avoidance46.3−12.7#222/2681/1
§ 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 GPT-6 Sol, left for the other.
§ 3 · Sources
Where the numbers come from
11 publications, 37 figures. Every one links to the page it was read from.
LMArena Text 1457
GDPval-AA 1516
AA-Briefcase 1476
Terminal-Bench 50.3% (Codex)
ARC-AGI-2 89.6%
LiveBench 1.0 79.3LiveBench 1.0 · Reasoning 88.7LiveBench 1.0 · Coding 81.8LiveBench 1.0 · Agentic Coding 52.9LiveBench 1.0 · Mathematics 96.4LiveBench 1.0 · Data Analysis 81.2LiveBench 1.0 · Language 85.3LiveBench 1.0 · Instruction Following 68.6
SimpleBench 73.1%
Vals · Legal Research Bench 28.85%Vals · Finance Agent 49.05%Vals · MedCode 47.07%Vals · MedScribe 82.03%Vals · BioMysteryBench 74.81%Vals · CyberBench 77.98%Vals · SRE Bench 30.53%Vals · Vibe Code Bench 87.82% (OpenHands)Vals · Code Migration 57.2%Vals · ProofBench 83%
ParseBench · Overall 68.85%ParseBench · Tables 92.65%ParseBench · Charts 48.05%ParseBench · Content faithfulness 87.54%ParseBench · Semantic formatting 67.91%ParseBench · Visual grounding 48.12%
Enkrypt · Jailbreak risk 12.2%Enkrypt · Harmful content risk 0%Enkrypt · CBRN risk 31.8%Enkrypt · Toxicity risk 8.7%Enkrypt · Bias risk 71.6%Enkrypt · Insecure code risk 18.2%
Vectara · Factual consistency 93.5%
Badge
[](https://publicai.io/model-index/m/gpt-6-sol)