‹ PublicAI Index
The LLM benchmark aggregator.
GPT-6 Astra
OpenAI
Strongest in IT operations (#1 of 33), weakest in Safety (#155 of 329). Above par in 27 of 29 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 5.5 and Claude Fable 5.1 and ahead of Gemini 4 Argon and Claude Sonnet 5.5.
§ 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.
Core abilities65.3leads#2/1993/4
Data analysis62.4−2.8#2/311/1
General intelligence67.8−1.6#3/1782/3
Language64.9−1.7#4/621/1
Instruction following63.5−10.3#7/621/1
Reasoning64.2−1.2#4/1955/5
Mathematics63.3−0.5#4/1522/2
Reasoning65.9−3.1#4/1473/3
Coding62.7−3.5#5/1794/6
Agentic coding63.2−3.7#5/1734/5
Agents63.8−2.6#6/2983/6
Long-horizon work70.9−3.1#2/311/1
Web research59.6−4.3#4/351/2
Computer use62−6.1#10/681/2
Knowledge work65.2−5.5#20/2082/2
Knowledge57.4−5.1#23/1971/3
Document parsing57.9−5.4#22/831/1
Professional54.7−7.7#33/1942/2
IT operations73.8leads#1/331/1
Biology research64.2−6.5#4/311/1
Document design60.8−7.9#5/311/1
Medical58−8.5#21/1491/1
Cybersecurity28−35.9#44/471/1
Finance53.7−8.7#68/1621/1
Legal48.8−17.1#100/1731/1
Human preference63.8−5#35/3541/1
Human preference63.8−5#35/3541/1
Safety51−10.4#155/3291/3
Factual grounding53.1−17.6#44/1041/1
§ 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 Astra, left for the other.
§ 3 · Sources
Where the numbers come from
12 publications, 41 figures. Every one links to the page it was read from.
LMArena Text 1474
Artificial Analysis Intelligence Index 57
GDPval-AA 1586
AA-Briefcase 1569
Terminal-Bench 60.3% (Codex)
ARC-AGI-2 95%
LiveBench 84.5LiveBench · Agentic 75.6LiveBench · Reasoning 92.7LiveBench · Coding 87.5LiveBench · Browser Use 91.3LiveBench · Data Analysis 83LiveBench · Research 90.6LiveBench · Design 80
LiveBench 1.0 82.2LiveBench 1.0 · Reasoning 92.7LiveBench 1.0 · Coding 80.4LiveBench 1.0 · Agentic Coding 57.3LiveBench 1.0 · Mathematics 96.8LiveBench 1.0 · Data Analysis 83LiveBench 1.0 · Language 89.4LiveBench 1.0 · Instruction Following 75.6
SimpleBench 83.6%
Vals · Legal Research Bench 39.42%Vals · Harvey Legal Agent Benchmark 7.08%Vals · Finance Agent 53.54%Vals · MedCode 48.49%Vals · MedScribe 87.91%Vals · BioMysteryBench 79.26%Vals · CyberBench 41.07%Vals · SRE Bench 56.87%Vals · Vibe Code Bench 89.59% (OpenHands)Vals · Code Migration 67.74%Vals · ProofBench 99%
ParseBench · Overall 69.81%ParseBench · Tables 93.18%ParseBench · Charts 35.08%ParseBench · Content faithfulness 89.21%ParseBench · Semantic formatting 75.52%ParseBench · Visual grounding 56.04%
Vectara · Factual consistency 91.3%
Badge
[](https://publicai.io/model-index/m/gpt-6-astra)