‹ PublicAI Index
The LLM benchmark aggregator.
GPT-6 Luna
OpenAI
Strongest in Cybersecurity (#5 of 47), weakest in Toxicity avoidance (#208 of 268). Above par in 17 of 33 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5 and Claude Opus 5.5 and ahead of Claude Sonnet 4 and Inkling.
§ 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.
Knowledge52.8−9.7#70/1971/3
Document parsing53−10.3#40/831/1
Coding51.8−14.4#74/1794/6
Agentic coding51.9−15#70/1734/5
Human preference60.6−8.2#81/3541/1
Human preference60.6−8.2#81/3541/1
Safety53.4−8#87/3291/3
Secure code63.6−2.6#20/2701/1
Fairness60.2−9.8#41/2921/2
Harm refusal50.1−11.4#152/2921/2
Jailbreak resistance50−15#176/2681/1
Toxicity avoidance48.6−10.4#208/2681/1
Reasoning49.8−15.6#103/1954/5
Reasoning48.5−20.5#72/1472/3
Mathematics50.7−13.1#83/1522/2
Professional49−13.4#112/1942/2
Cybersecurity60.2−3.7#5/471/1
Document design54.7−14#13/311/1
IT operations41.1−32.7#24/331/1
Biology research38.2−32.5#27/311/1
Medical53.8−12.7#58/1491/1
Finance51.4−11#87/1621/1
Legal43.8−22.1#131/1731/1
Agents49.3−17.1#141/2983/6
Long-horizon work43.1−30.9#20/311/1
Computer use54.4−13.7#25/681/2
Web research30−33.9#34/351/2
Knowledge work60.9−9.8#43/2082/2
Core abilities44.1−21.2#156/1993/4
Data analysis50−15.2#20/311/1
Language44.4−22.2#48/621/1
Instruction following28.6−45.2#59/621/1
General intelligence48.8−20.6#101/1782/3
§ 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 Luna, left for the other.
§ 3 · Sources
Where the numbers come from
11 publications, 45 figures. Every one links to the page it was read from.
LMArena Text 1441
Artificial Analysis Intelligence Index 28
GDPval-AA 1444
AA-Briefcase 1337
Terminal-Bench 15.2% (Codex)
ARC-AGI-2 59.3%
LiveBench 60.4LiveBench · Agentic 24.7LiveBench · Reasoning 81.8LiveBench · Coding 70.8LiveBench · Browser Use 70.3LiveBench · Data Analysis 73.4LiveBench · Research 48.5LiveBench · Design 76.1
LiveBench 1.0 72LiveBench 1.0 · Reasoning 81.8LiveBench 1.0 · Coding 79LiveBench 1.0 · Agentic Coding 51.2LiveBench 1.0 · Mathematics 89.1LiveBench 1.0 · Data Analysis 73.4LiveBench 1.0 · Language 73.8LiveBench 1.0 · Instruction Following 55.9
Vals · Legal Research Bench 30.29%Vals · Harvey Legal Agent Benchmark 4.17%Vals · Finance Agent 49.87%Vals · MedCode 44.69%Vals · MedScribe 83.71%Vals · BioMysteryBench 61.48%Vals · CyberBench 76.25%Vals · SRE Bench 2.67%Vals · Vibe Code Bench 81.65% (OpenHands)Vals · Code Migration 42.55%Vals · ProofBench 64%
ParseBench · Overall 64.13%ParseBench · Tables 87.44%ParseBench · Charts 45.71%ParseBench · Content faithfulness 84.21%ParseBench · Semantic formatting 61.81%ParseBench · Visual grounding 41.48%
Enkrypt · Jailbreak risk 12.8%Enkrypt · Harmful content risk 0%Enkrypt · CBRN risk 33%Enkrypt · Toxicity risk 7.1%Enkrypt · Bias risk 62%Enkrypt · Insecure code risk 5.3%
Badge
[](https://publicai.io/model-index/m/gpt-6-luna)