‹ PublicAI Index
The LLM benchmark aggregator.
Claude Opus 4.5
Anthropic
Strongest in Tool use (#1 of 81), weakest in Safety (#202 of 329). Above par in 17 of 22 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5 and Claude Fable 5.1 and ahead of GPT-5.1 and GPT-5.2.
§ 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.
Agents60−6.4#12/2981/6
Tool use74leads#1/811/1
Professional56.6−5.8#19/1941/2
Medical58.4−8.1#19/1491/1
Finance57.2−5.2#28/1621/1
Legal55.7−10.2#39/1731/1
Human preference63.8−5#33/3541/1
Human preference63.8−5#33/3541/1
Reasoning54.9−10.5#46/1954/5
Mathematics55.9−7.9#44/1522/2
Science56.8−6.7#45/1221/1
Reasoning52.7−16.3#54/1472/3
Knowledge53.1−9.4#59/1971/3
Academic knowledge57.5−6.4#31/1231/1
Core abilities53.6−11.7#61/1992/4
General intelligence63.5−5.9#14/1781/3
Language54.2−12.4#25/621/1
Instruction following40.3−33.5#50/621/1
Coding43.9−22.3#136/1792/6
Agentic coding43.6−23.3#131/1732/5
Safety49.2−12.2#202/3291/3
Factual grounding47.6−23.1#70/1041/1
§ 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 Claude Opus 4.5, left for the other.
§ 3 · Sources
Where the numbers come from
8 publications, 27 figures. Every one links to the page it was read from.
LMArena Text 1474
ARC-AGI-2 37.6%
LiveBench 1.0 72.6LiveBench 1.0 · Reasoning 80.1LiveBench 1.0 · Coding 79.7LiveBench 1.0 · Agentic Coding 39.7LiveBench 1.0 · Mathematics 90.4LiveBench 1.0 · Data Analysis 74.4LiveBench 1.0 · Language 81.3LiveBench 1.0 · Instruction Following 62.5
BFCL v4 77.47%
Kagi LLM Benchmark 80.2%
SimpleBench 62%
Vals · CaseLaw 62.59%Vals · LegalBench 84.6%Vals · CorpFin 65.07%Vals · TaxEval 74.86%Vals · MortgageTax 67.69%Vals · MedQA 95.88%Vals · MedCode 49.16%Vals · MedScribe 85.32%Vals · SWE-bench Verified 76.4% (Mini-SWE-agent)Vals · Vibe Code Bench 20.63% (OpenHands)Vals · GPQA Diamond 85.86%Vals · MMLU Pro 87.26%Vals · AIME 95.42%
Vectara · Factual consistency 89.1%
Badge
[](https://publicai.io/model-index/m/claude-opus-4-5)