‹ PublicAI Index
The LLM benchmark aggregator.
Claude Opus 4.8
Anthropic
Strongest in Safety (#6 of 327, on 1 of its 3 boards), weakest in Secure code (#122 of 268). Above par in 27 of 27 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5 and Claude Fable 5.1 and ahead of Gemini 3.1 Pro and GPT-6 Sol.
§ 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.
Safety59.8−1.6#6/3271/3
Fairness69.1−0.9#13/2901/2
Harm refusal59.2−2.3#16/2901/2
Jailbreak resistance61.7−3.3#27/2661/1
Toxicity avoidance57.9−1.1#31/2661/1
Secure code54−12.2#122/2681/1
Professional57.6−4.8#12/1941/2
Finance58.7−3.7#9/1621/1
Medical59.8−6.7#9/1491/1
Legal56.4−9.5#31/1731/1
Core abilities59.3−6.3#13/2002/4
General intelligence67.6−1.4#6/1801/3
Instruction following57.1−16.7#19/621/1
Language52.1−14.5#31/621/1
Reasoning58.1−7.3#19/1954/5
Science61.4−2.1#16/1221/1
Reasoning57.4−11.6#34/1472/3
Mathematics56.7−7.1#38/1521/2
Human preference64.5−4.3#24/3541/1
Human preference64.5−4.3#24/3541/1
Coding55.4−10.7#30/1793/6
Agentic coding55.6−11.1#32/1733/5
Agents56.8−9.5#38/2982/6
Knowledge work60.9−9.8#44/2082/2
Knowledge54.6−7.9#41/1972/3
Academic knowledge60.4−3.5#9/1231/1
Document parsing53.8−9.5#38/831/1
§ 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 4.8, left for the other.
§ 3 · Sources
Where the numbers come from
11 publications, 40 figures. Every one links to the page it was read from.
LMArena Text 1482
GDPval-AA 1462
AA-Briefcase 1320
Terminal-Bench 23.6% (Claude Code)
ARC-AGI-2 72.1%
LiveBench 1.0 76.2LiveBench 1.0 · Reasoning 89.2LiveBench 1.0 · Coding 81.8LiveBench 1.0 · Agentic Coding 50.5LiveBench 1.0 · Mathematics 94.3LiveBench 1.0 · Data Analysis 66LiveBench 1.0 · Language 79.7LiveBench 1.0 · Instruction Following 72
Kagi LLM Benchmark 88.8%
SimpleBench 64.8%
Vals · Legal Research Bench 43.75%Vals · LegalBench 83.57%Vals · Finance Agent 53.92%Vals · CorpFin 66.71%Vals · TaxEval 75.63%Vals · MortgageTax 69.91%Vals · MedCode 53.22%Vals · MedScribe 85.75%Vals · SWE-bench Verified 88.6% (Mini-SWE-agent)Vals · Vibe Code Bench 82.72% (OpenHands)Vals · Code Migration 47.25%Vals · GPQA Diamond 92.42%Vals · MMLU Pro 89.58%
ParseBench · Overall 62.27%ParseBench · Tables 89.65%ParseBench · Charts 49.75%ParseBench · Content faithfulness 89.02%ParseBench · Semantic formatting 71.38%ParseBench · Visual grounding 11.53%
Enkrypt · Jailbreak risk 3%Enkrypt · Harmful content risk 0.6%Enkrypt · CBRN risk 8%Enkrypt · Toxicity risk 0.7%Enkrypt · Bias risk 48.1%Enkrypt · Insecure code risk 24.9%
Badge
[](https://publicai.io/model-index/m/claude-opus-4-8)