‹ PublicAI Index
The LLM benchmark aggregator.
Claude Opus 5.5
Anthropic
Strongest in Knowledge work (#1 of 208), weakest in Toxicity avoidance (#119 of 268). Above par in 32 of 34 scopes. Among the models it meets almost everywhere, it finishes behind none and ahead of Gemini 4 Argon and Claude Fable 5.1.
§ 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.
Knowledge62.5leads#1/1971/3
Document parsing63.3leads#1/831/1
Coding66.2leads#1/1794/6
Agentic coding66.9leads#1/1734/5
Agents65.9−0.5#2/2984/6
Knowledge work70.7leads#1/2082/2
Web research63.9leads#1/352/2
Long-horizon work69.4−4.6#4/311/1
Computer use61−7.1#14/681/2
Reasoning64.6−0.8#2/1955/5
Mathematics63.8leads#1/1522/2
Reasoning66.4−2.6#3/1473/3
Human preference66.8−2#4/3541/1
Human preference66.8−2#4/3541/1
Professional58−4.4#8/1942/2
Document design65.9−2.8#2/311/1
IT operations64.4−9.4#4/331/1
Biology research64.2−6.5#4/311/1
Medical60.5−6#8/1491/1
Finance57−5.4#30/1621/1
Cybersecurity41.4−22.5#39/471/1
Legal54−11.9#54/1731/1
Safety59.5−1.9#10/3292/3
Secure code66−0.2#2/2701/1
Fairness70leads#3/2921/2
Harm refusal58.2−3.3#23/2921/2
Jailbreak resistance59.9−5.1#48/2681/1
Factual grounding50.1−20.6#58/1041/1
Toxicity avoidance54.5−4.5#119/2681/1
Core abilities60.4−4.9#12/1993/4
General intelligence68.1−1.3#2/1782/3
Data analysis58.9−6.3#7/311/1
Language60.8−5.8#9/621/1
Instruction following45.9−27.9#39/621/1
§ 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 Claude Opus 5.5, left for the other.
§ 3 · Sources
Where the numbers come from
13 publications, 48 figures. Every one links to the page it was read from.
LMArena Text 1506
Artificial Analysis Intelligence Index 59
GDPval-AA 1866
AA-Briefcase 1801
Terminal-Bench 65.8% (Claude Code)
ARC-AGI-2 91.7%
LiveBench 83.4LiveBench · Agentic 72.8LiveBench · Reasoning 92.2LiveBench · Coding 87.1LiveBench · Browser Use 88.4LiveBench · Data Analysis 80.3LiveBench · Research 88.9LiveBench · Design 83.3
LiveBench 1.0 83.2LiveBench 1.0 · Reasoning 92.2LiveBench 1.0 · Coding 89.3LiveBench 1.0 · Agentic Coding 71.7LiveBench 1.0 · Mathematics 97.1LiveBench 1.0 · Data Analysis 80.3LiveBench 1.0 · Language 86.3LiveBench 1.0 · Instruction Following 65.7
SimpleBench 88.4%
Vals · Legal Research Bench 50.48%Vals · Harvey Legal Agent Benchmark 9.17%Vals · Finance Agent 58.59%Vals · MedCode 49.8%Vals · MedScribe 91.43%Vals · BioMysteryBench 79.26%Vals · CyberBench 55.36%Vals · SRE Bench 33.59%Vals · Web Search Index 57.54% (Parallel Fast)Vals · Vibe Code Bench 90.29% (OpenHands)Vals · Code Migration 66.65%Vals · ProofBench 100%
ParseBench · Overall 78.8%ParseBench · Tables 94.26%ParseBench · Charts 70.89%ParseBench · Content faithfulness 91.25%ParseBench · Semantic formatting 77.06%ParseBench · Visual grounding 60.56%
Enkrypt · Jailbreak risk 4.5%Enkrypt · Harmful content risk 1.7%Enkrypt · CBRN risk 10.2%Enkrypt · Toxicity risk 3.1%Enkrypt · Bias risk 25.6%Enkrypt · Insecure code risk 0.4%
Vectara · Factual consistency 90.1%
Badge
[](https://publicai.io/model-index/m/claude-opus-5-5)