evals.report
BenchmarksLabsCompareRun guidesIn the wild
BenchmarksOther

MASK (Model Alignment between Statements and Knowledge)

A human-collected honesty benchmark that first elicits a model's beliefs, then measures whether the model maintains truthful assertions when directly or indirectly pressured to lie, disentangling honesty from factual accuracy.

OtherHonesty scoreHigher is better
ScoresAboutRun this benchmark

What is MASK (Model Alignment between Statements and Knowledge)?

A human-collected honesty benchmark that first elicits a model's beliefs, then measures whether the model maintains truthful assertions when directly or indirectly pressured to lie, disentangling honesty from factual accuracy. evals.report tracks reported MASK (Model Alignment between Statements and Knowledge) scores with the model, source, status, date, and run caveats attached — official leaderboard scores, vendor-reported launches, and clearly labeled community runs.

Top reported MASK (Model Alignment between Statements and Knowledge) score: Claude Opus 4.6 — 96.28 (Honesty score).

ModelLabScore↓Source modelStatusDate
Claude Opus 4.6Anthropic96.28—VerifiedFeb 5, 2026Details
Claude Sonnet 4.5Anthropic96.13—VerifiedSep 29, 2025Details
Claude Sonnet 4Anthropic95.33—VerifiedMay 22, 2025Details
Claude Opus 4.1Anthropic94.20—VerifiedAug 5, 2025Details
Claude Opus 4.5Anthropic92.53—VerifiedNov 24, 2025Details
GPT-OSS-120BOpenOpenAI92.00—VerifiedAug 5, 2025Details
GPT-5.4 ProOpenAI91.73—VerifiedMar 5, 2026Details
GPT-5.4OpenAI89.67—VerifiedMar 5, 2026Details
Claude Opus 4Anthropic87.87—VerifiedMay 22, 2025Details
GPT-5.2OpenAI86.67—VerifiedDec 11, 2025Details
GPT-5.1OpenAI86.33—VerifiedNov 12, 2025Details
o3OpenAI84.47—VerifiedApr 16, 2025Details
GPT-5 miniOpenAI82.60—VerifiedAug 7, 2025Details
OpenAI o3-proOpenAI82.50—VerifiedJun 10, 2025Details
Claude 3.7 SonnetAnthropic82.13—VerifiedFeb 24, 2025Details
GPT-5OpenAI79.33—VerifiedAug 7, 2025Details
o4-miniOpenAI78.60—VerifiedApr 16, 2025Details
Claude 3.5 SonnetAnthropic72.33—VerifiedJun 20, 2024Details
Kimi K2.5OpenMoonshot AI70.47—VerifiedJan 27, 2026Details
Llama 3.1 405BOpenMeta61.40—VerifiedJul 23, 2024Details
GPT-4oOpenAI60.07—VerifiedMay 13, 2024Details
DeepSeek R1OpenDeepSeek57.32—VerifiedJan 20, 2025Details
Gemini 2.5 ProGoogle DeepMind55.67—VerifiedMar 25, 2025Details
GPT-4.1OpenAI51.13—VerifiedApr 14, 2025Details
Llama 4 MaverickOpenMeta49.73—VerifiedApr 5, 2025Details
Gemini 2.5 FlashGoogle DeepMind49.13—VerifiedApr 17, 2025Details
Gemini 2.0 FlashGoogle DeepMind49.07—VerifiedDec 11, 2024Details
Mistral LargeMistral AI47.53—VerifiedFeb 26, 2024Details
Kimi K2 InstructOpenMoonshot AI46.67—VerifiedJul 11, 2025Details
DeepSeek V3.1OpenDeepSeek46.27—VerifiedAug 21, 2025Details
DeepSeek V3 0324OpenDeepSeek44.53—VerifiedMar 24, 2025Details
Gemini 3 ProGoogle DeepMind42.60—VerifiedNov 18, 2025Details
Gemini 3.1 Pro PreviewGoogle DeepMind42.40—VerifiedFeb 19, 2026Details

Each row reports the model’s Honesty score on MASK (Model Alignment between Statements and Knowledge). Click a row for the full run context.