← Evals

Evaluation profile

Opposite-Narrator Sycophancy

1sub-evals
0.344%total index weight
1components

Within-component eval weight: Truthfulness 2.29%.

Model score (lower is better)Predicted score

About this eval

Narrator-following contradiction when the same dispute is presented from opposite affective first-person perspectives.

Included in the behavior ranking.

Sub-evals

MeasureComponentDirectionTotal index weightWithin-component weight
sycophancy_rate_pctopposite-narrator-sycophancy/opposite-narrator-sycophancy.csv:sycophancy_rate_pctMeasures how often a model sides with both opposing speakers when each tells the same dispute from their own affective first-person perspective.truthfulness_honesty:1.000opposite-narrator-sycophancyLower is better0.344%Truthfulness 2.29%

sycophancy_rate_pct

Measures how often a model sides with both opposing speakers when each tells the same dispute from their own affective first-person perspective.

RankModelValueRelative performanceProvenance
1gpt-5.6-terra0official
1grok-4.50official
3claude-fable-50.5official
3gemini-3.6-flash0.5official
3hy30.5official
6gemini-3.5-flash-lite1official
6gpt-5.6-luna1official
8qwen3.7-max1.5official
9ernie-5.12official
10qwen3.7-flash2.5official
11mimo-v2.5-pro2.6official
12gpt-5.6-sol3official
13inkling3.5official
14minimax-m33.6official
15kimi-k34.5official
16gemma-4-31b-it4.6official
17deepseek-v4-pro5.1official
17qwen3.7-plus5.1official
19deepseek-v4-flash5.6official
20claude-sonnet-59.1official
21glm-5.212.6official
22seed-2.1-pro14.6official
23trinity-large18.9official
24mistral-medium-3.522.4official