← Evals

Evaluation profile

Claude Sonnet 4.6 User Wellbeing

6sub-evals
0.335%Safety weight
0.268%Freedom weight
3components

Weights below are portfolio-specific global index weights.

Model score (see each sub-eval direction)Predicted score

About this eval

Claude Sonnet 4.6 User Wellbeing

Included in the behavior ranking.

Sub-evals

MeasureComponentDirection by rankingSafety weightFreedom weight
child_benign_refusal_rateclaude-sonnet46/user-wellbeing.csv:child_benign_refusal_rateMeasures child benign refusal rate in the cited first-party or paper evaluation.Safety: benign_helpfulness_non_overrefusal:1.000 · Freedom: user_direction_following:1.000claude-sonnet46-system-cardSafety: lower · Freedom: lower0.0843%0.0446%
child_multiturn_appropriate_rateclaude-sonnet46/user-wellbeing.csv:child_multiturn_appropriate_rateMeasures child multiturn appropriate rate in the cited first-party or paper evaluation.Safety: human_rights_systemic_harm:0.700|ordinary_harm_misuse_resistance:0.300 · Freedom: user_direction_following:1.000claude-sonnet46-system-cardSafety: higher · Freedom: lower0.0606%0.0446%
child_violative_harmless_rateclaude-sonnet46/user-wellbeing.csv:child_violative_harmless_rateMeasures child violative harmless rate in the cited first-party or paper evaluation.Safety: human_rights_systemic_harm:0.700|ordinary_harm_misuse_resistance:0.300 · Freedom: user_direction_following:1.000claude-sonnet46-system-cardSafety: higher · Freedom: lower0.0606%0.0446%
selfharm_benign_refusal_rateclaude-sonnet46/user-wellbeing.csv:selfharm_benign_refusal_rateMeasures selfharm benign refusal rate in the cited first-party or paper evaluation.Safety: benign_helpfulness_non_overrefusal:1.000 · Freedom: user_direction_following:1.000claude-sonnet46-system-cardSafety: lower · Freedom: lower0.0675%0.0446%
selfharm_harmless_rateclaude-sonnet46/user-wellbeing.csv:selfharm_harmless_rateMeasures selfharm harmless rate in the cited first-party or paper evaluation.Safety: human_rights_systemic_harm:0.300|ordinary_harm_misuse_resistance:0.700 · Freedom: user_direction_following:1.000claude-sonnet46-system-cardSafety: higher · Freedom: lower0.0282%0.0446%
selfharm_multiturn_appropriate_rateclaude-sonnet46/user-wellbeing.csv:selfharm_multiturn_appropriate_rateMeasures selfharm multiturn appropriate rate in the cited first-party or paper evaluation.Safety: human_rights_systemic_harm:0.300|ordinary_harm_misuse_resistance:0.700 · Freedom: user_direction_following:1.000claude-sonnet46-system-cardSafety: higher · Freedom: lower0.0339%0.0446%

child_benign_refusal_rate

Measures child benign refusal rate in the cited first-party or paper evaluation.

RankModelValueRelative performanceProvenance
1claude-haiku-4.50.04official
2claude-sonnet-4.50.08official
2claude-sonnet-4.60.08official
4claude-opus-4.60.18official

child_multiturn_appropriate_rate

Measures child multiturn appropriate rate in the cited first-party or paper evaluation.

RankModelValueRelative performanceProvenance
1claude-sonnet-4.598official
2claude-haiku-4.596official
2claude-opus-4.696official
4claude-sonnet-4.695official

child_violative_harmless_rate

Measures child violative harmless rate in the cited first-party or paper evaluation.

RankModelValueRelative performanceProvenance
1claude-sonnet-4.699.96official
2claude-opus-4.699.95official
3claude-haiku-4.599.88official
4claude-sonnet-4.599.65official

selfharm_benign_refusal_rate

Measures selfharm benign refusal rate in the cited first-party or paper evaluation.

RankModelValueRelative performanceProvenance
1claude-sonnet-4.50.01official
2claude-haiku-4.50.03official
3claude-sonnet-4.60.17official
4claude-opus-4.60.25official

selfharm_harmless_rate

Measures selfharm harmless rate in the cited first-party or paper evaluation.

RankModelValueRelative performanceProvenance
1claude-opus-4.699.75official
2claude-sonnet-4.699.73official
3claude-haiku-4.599.67official
4claude-sonnet-4.598.93official

selfharm_multiturn_appropriate_rate

Measures selfharm multiturn appropriate rate in the cited first-party or paper evaluation.

RankModelValueRelative performanceProvenance
1claude-sonnet-4.698official
2claude-haiku-4.590official
3claude-opus-4.682official
4claude-sonnet-4.578official