← Evals

Evaluation profile

KIDBench Implicit Child Cue

1sub-evals
0.685%total index weight
3components

Within-component eval weight: Human rights 1.24% · Benign helpfulness 4.39% · Misuse resistance 0.604%.

Model score (higher is better)Predicted score

About this eval

Child-facing safety and developmental appropriateness under implicit child cues.

Included in the behavior ranking.

Sub-evals

MeasureComponentDirectionTotal index weightWithin-component weight
implicit_child_cue_total_meankidbench/kidbench.csv:implicit_child_cue_total_meanMeasures safe and developmentally appropriate responses when child status is implied rather than explicit.human_rights_systemic_harm:0.200|benign_helpfulness_non_overrefusal:0.400|ordinary_harm_misuse_resistance:0.400kidbench-implicit-child-cueHigher is better0.685%Human rights 1.24% · Benign helpfulness 4.39% · Misuse resistance 0.604%

implicit_child_cue_total_mean

Measures safe and developmentally appropriate responses when child status is implied rather than explicit.

RankModelValueRelative performanceProvenance
1deepseek-v4-flash4.43official
2qwen3.6-27b4.29official
3gemini-3.1-flash-lite4.27official
4gemma-4-31b-it4.14official
5claude-haiku-4.54official
6gpt-5-mini3.99official
7gemma-3-12b3.98official
8qwen3.5-4b3.9official
9llama-3.3-70b-instruct3.76official
10qwen3-8b3.59official
11gemma-3-4b3.52official
12llama-3.1-8b-instruct3.11official
13llama-3.2-3b3.03official