‹ PublicAI Index
The LLM benchmark aggregator.
Claude Opus 4.7
Anthropic
Strongest in Harm refusal (#2 of 292, on 1 of its 2 boards), weakest in Fairness (#100 of 292). Above par in 25 of 28 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5 and Claude Fable 5.1 and ahead of Grok 4.7 and Gemini 3.1 Pro.
§ 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.
Human preference66.4−2.4#8/3541/1
Human preference66.4−2.4#8/3541/1
Professional58−4.4#9/1941/2
Medical59.5−7#13/1491/1
Legal59.2−6.7#15/1731/1
Finance57.9−4.5#17/1621/1
Reasoning58−7.4#21/1954/5
Mathematics58.2−5.6#20/1522/2
Science59.8−3.7#26/1221/1
Reasoning56.9−12.1#38/1472/3
Safety56.4−5#35/3292/3
Harm refusal60.8−0.7#2/2921/2
Jailbreak resistance63.3−1.7#9/2681/1
Toxicity avoidance56.8−2.2#47/2681/1
Factual grounding44.8−25.9#77/1041/1
Secure code57.5−8.7#90/2701/1
Fairness50.9−19.1#100/2921/2
Knowledge55.3−7.2#38/1972/3
Academic knowledge60.8−3.1#8/1231/1
Document parsing54.6−8.7#36/831/1
Coding54.3−11.9#45/1792/6
Agentic coding54.5−12.4#44/1732/5
Agents55.7−10.7#50/2982/6
Knowledge work59.1−11.6#52/2082/2
Core abilities54.5−10.8#52/1992/4
General intelligence63.7−5.7#12/1781/3
Language49.8−16.8#36/621/1
Instruction following47.7−26.1#36/621/1
§ 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 4.7, left for the other.
§ 3 · Sources
Where the numbers come from
11 publications, 42 figures. Every one links to the page it was read from.
LMArena Text 1501
GDPval-AA 1362
AA-Briefcase 1252
ARC-AGI-2 75.8%
LiveBench 1.0 76.5LiveBench 1.0 · Reasoning 87.2LiveBench 1.0 · Coding 82.1LiveBench 1.0 · Agentic Coding 50.7LiveBench 1.0 · Mathematics 92.9LiveBench 1.0 · Data Analysis 78.3LiveBench 1.0 · Language 77.9LiveBench 1.0 · Instruction Following 66.7
Kagi LLM Benchmark 80.7%
SimpleBench 61.7%
Vals · Legal Research Bench 38.46%Vals · CaseLaw 68.38%Vals · LegalBench 85.25%Vals · Finance Agent 51.51%Vals · CorpFin 66.08%Vals · TaxEval 75.27%Vals · MortgageTax 70.27%Vals · MedCode 54.86%Vals · MedScribe 82.95%Vals · SWE-bench Verified 82% (Mini-SWE-agent)Vals · Vibe Code Bench 71% (OpenHands)Vals · Code Migration 43.88%Vals · GPQA Diamond 90.15%Vals · MMLU Pro 89.87%Vals · AIME 96.25%
ParseBench · Overall 63.34%ParseBench · Tables 87.17%ParseBench · Charts 55.84%ParseBench · Content faithfulness 90.26%ParseBench · Semantic formatting 69.42%ParseBench · Visual grounding 13.99%
Enkrypt · Jailbreak risk 1.7%Enkrypt · Harmful content risk 0.6%Enkrypt · CBRN risk 3.7%Enkrypt · Toxicity risk 1.5%Enkrypt · Bias risk 76.5%Enkrypt · Insecure code risk 17.8%
Vectara · Factual consistency 88%
Badge
[](https://publicai.io/model-index/m/claude-opus-4-7)