← Evals

Evaluation profile

RealityTest — Text AI-Identity Disclosure

1sub-evals
0.289%total index weight
1components

Within-component eval weight: Truthfulness 1.93%.

Model score (higher is better)Predicted score

About this eval

Behavioral disclosure of AI identity when directly queried.

Included in the behavior ranking.

Sub-evals

MeasureComponentDirectionTotal index weightWithin-component weight
disclosure_probabilityrealitytest/realitytest-text.csv:disclosure_probabilityMeasures whether a model says it is an AI when directly asked in text.truthfulness_honesty:1.000realitytest-textHigher is better0.289%Truthfulness 1.93%

disclosure_probability

Measures whether a model says it is an AI when directly asked in text.

RankModelValueRelative performanceProvenance
1claude-haiku-4.50.923official
2gpt-5.10.856official
3claude-opus-4.60.66official
4kimi-k20.552official
5voxtral-small-24b-25070.541official
6claude-sonnet-40.515official
7grok-40.413official
8gemma-4-31b-it0.313official
9gemini-3.1-pro-preview0.305official
10o4-mini0.283official
11deepseek-r10.234official
12gemini-3-flash-preview0.227official
13deepseek-v3.20.147official
14gemini-2.5-pro0.128official
15gpt-4o0.126official
16llama-3.3-70b-instruct0.125official
17mistral-large-30.078official