‹ PublicAI Index
The LLM benchmark aggregator.
Claude Opus 4.6
Anthropic
Strongest in Human preference (#6 of 354), weakest in Secure code (#196 of 270). Above par in 20 of 24 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5 and Gemini 4 Argon and ahead of GPT-5 and DeepSeek V4.1 Flash.
§ 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 preference66.6−2.2#6/3541/1
Human preference66.6−2.2#6/3541/1
Professional57.1−5.3#14/1941/2
Finance58.7−3.7#10/1621/1
Medical58.5−8#18/1491/1
Legal55.7−10.2#40/1731/1
Reasoning57−8.4#30/1954/5
Science59.5−4#28/1221/1
Reasoning57.9−11.1#31/1472/3
Mathematics55−8.8#55/1522/2
Core abilities55.4−9.9#41/1992/4
General intelligence65.2−4.2#9/1781/3
Language56.9−9.7#18/621/1
Instruction following41.7−32.1#47/621/1
Coding50.3−15.9#84/1792/6
Agentic coding50.3−16.6#83/1732/5
Knowledge50.5−12#105/1972/3
Academic knowledge59.9−4#15/1231/1
Document parsing49−14.3#51/831/1
Safety51.9−9.5#133/3292/3
Harm refusal58.9−2.6#18/2921/2
Factual grounding44.3−26.4#83/1041/1
Fairness50.4−19.6#109/2921/2
Secure code44.5−21.7#196/2701/1
§ 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 4.6, left for the other.
§ 3 · Sources
Where the numbers come from
9 publications, 36 figures. Every one links to the page it was read from.
LMArena Text 1503
ARC-AGI-2 68.8%
LiveBench 1.0 74.5LiveBench 1.0 · Reasoning 88.7LiveBench 1.0 · Coding 78.2LiveBench 1.0 · Agentic Coding 49LiveBench 1.0 · Mathematics 89.3LiveBench 1.0 · Data Analysis 69.9LiveBench 1.0 · Language 83.3LiveBench 1.0 · Instruction Following 63.3
Kagi LLM Benchmark 83.6%
SimpleBench 67.6%
Vals · CaseLaw 62.06%Vals · LegalBench 85.3%Vals · CorpFin 67.02%Vals · TaxEval 75.96%Vals · MortgageTax 68.52%Vals · MedQA 95.41%Vals · MedCode 49.13%Vals · MedScribe 86.13%Vals · SWE-bench Verified 78.2% (Mini-SWE-agent)Vals · Vibe Code Bench 53.5% (OpenHands)Vals · GPQA Diamond 89.65%Vals · MMLU Pro 89.11%Vals · AIME 95.63%
ParseBench · Overall 53.8%ParseBench · Tables 86.25%ParseBench · Charts 15.03%ParseBench · Content faithfulness 89.23%ParseBench · Semantic formatting 62.04%ParseBench · Visual grounding 16.44%
Enkrypt · Harmful content risk 0.6%Enkrypt · CBRN risk 8.7%Enkrypt · Bias risk 77.3%Enkrypt · Insecure code risk 44.4%
Vectara · Factual consistency 87.8%
Badge
[](https://publicai.io/model-index/m/claude-opus-4-6)