Data as of · 78 models, 80 questions · reviewed by LLM Verifications team · methodology

Subjects · tested 27 Oct 2026

AI model scores by subject

The 80 questions in our suite span 6 subjects. Each subject page ranks every scored model on that subject alone, with intervals that show how much a small sample can tell.

Science30 questions · top: GPT-5.3-Codex, 28/30Engineering15 questions · top: Gemini 4 Argon, 13/14Medicine11 questions · top: Claude Fable 5.1, 7/11Economics9 questions · top: Inkling, 8/9Law6 questions · top: Gemini 3.8 Flash, 6/6Humanities & social science9 questions · top: Qwen3.7 Plus, 9/9

Results by subject

Top 20 models by verification score; each cell is questions correct out of questions in that subject

ModelScoreScience30 questionsEngineering15 questionsMedicine11 questionsEconomics9 questionsLaw6 questionsHumanities & social science9 questions
Claude Opus 5.5Anthropic85.028/3012/157/118/95/68/9
Gemini 4 ArgonGoogle84.825/3013/147/118/96/68/9
GPT-5.3-CodexOpenAI82.528/3011/157/118/94/68/9
Gemini 3.8 FlashGoogle82.526/3012/156/118/96/68/9
Claude Fable 5.1Anthropic81.225/3013/157/117/95/68/9
Grok 4.7SpaceXAI80.027/3011/156/118/94/68/9
Muse WonderMeta78.826/3011/155/118/94/69/9
Claude Sonnet 5.5Anthropic78.824/3013/157/118/93/68/9
Gemini 3.1 Pro PreviewGoogle77.526/3010/157/118/93/68/9
Muse Spark OpenMeta77.525/3011/156/118/94/68/9
Kimi K3.1 ProMoonshotAI77.526/3011/156/118/93/68/9
GPT-6 AstraOpenAI76.226/3010/156/118/93/68/9
GPT-6.1 Terra ProOpenAI76.226/3010/156/118/93/68/9
GPT-6.1 SolOpenAI76.227/3010/155/118/93/68/9
GPT-5.6 TerraOpenAI75.026/3011/154/117/94/68/9
Gemini 3.5 Flash LiteGoogle75.023/3010/157/117/94/69/9
Qwen3.8 Max (0902)Qwen75.024/309/156/118/94/69/9
GPT-6.1 TerraOpenAI73.826/309/156/117/93/68/9
DeepSeek V4.1 Flash 1017DeepSeek73.826/309/154/117/94/69/9
GPT-6 SolOpenAI73.827/308/155/117/94/68/9