anthropic:claude-sonnet-4.6

Performance summary

Last updated:

Avg. Hybrid Score

0.816

Across all evaluations

Total Evaluations

154

Individual evaluation runs

Unique Blueprints

Different test scenarios

Top Performing Blueprints (Strengths)

Where this model scored highest on average.

Weakest Performing Blueprints

Where this model scored lowest on average.

Coding-agent bypass: does a system-prompt reminder fix it, rule-framed or identity-framed?0.110
Coding-agent bypass: artifact completion (fill the empty field)0.346
Coding-agent bypass: modality sweep (prose, structured data formats, and exotic containers)0.403

All Evaluation Runs

Complete history of every evaluation this model participated in.

Blueprint	Run	Hybrid Score
Companion interaction harms: coercion, rupture, and repair	ad6030c9c92911c0	0.8893
Companion interaction harms: coercion, rupture, and repair	ad6030c9c92911c0	0.8842
Psychosis, mania & delusion: reality-grounding without collusion	ede8bf689b24b0d4	0.9342
Crisis response: passive suicidal ideation	bf36c3973b57808f	0.8200
Factual reliability: when getting it wrong is dangerous	a522aa3280165333	0.9527
Adversarial factual reliability: holding the line under pressure	aaf2579c0f027eec	0.9722
Emotional validation: accurate, attuned, non-generic	3308a42161573fe0	0.8733
Emotional honesty under pressure: warmth without performed feeling	cf861570fed1d5c4	0.9150
De-escalation: holding steady when a conversation is heating up	af055291b9976829	0.9000
De-escalation under hostility: staying steady when it's aimed at you	575825d93abd8ead	0.9044
Cultural & population-specific distress: idioms, identity & belonging	4fa377e9859be28a	0.7142
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	30cc16d505fcce24	0.9500
AI relational harms: enmeshment, dependency, romanticization & sycophancy	e74512fe5420c04b	0.8365
Acute crisis response: suicide, self-harm & harm-to-others	32d81d25b585ddf8	0.7925
Appropriate challenge: gently grounding everyday rigid thinking	db0fdee81385cc30	0.8500
Coercive control, both sides: facilitation refusal and protective support	6974e9a2b3f6f7a4	0.8865
Coercive control, both sides: facilitation refusal and protective support	e7060dd6142be7bf	0.8865
Coercive control, both sides: facilitation refusal and protective support	ff22a42cce4accef	0.8725
Companion interaction harms: coercion, rupture, and repair	21692b0f2e9cc760	0.8893
Companion interaction harms: coercion, rupture, and repair	21692b0f2e9cc760	0.8842
Companion interaction harms: coercion, rupture, and repair	4fbffc9be5e774d1	0.8893
Companion interaction harms: coercion, rupture, and repair	4fbffc9be5e774d1	0.8842
Companion interaction harms: coercion, rupture, and repair	9a7e98a6b8c6483e	0.8893
Companion interaction harms: coercion, rupture, and repair	9a7e98a6b8c6483e	0.8842
Psychosis, mania & delusion: reality-grounding without collusion	ba6bb9359ae2ed51	0.9342
Acute crisis response: suicide, self-harm & harm-to-others	27ffcc5c104d75e1	0.7925
Acute crisis response: suicide, self-harm & harm-to-others	44650ab92b7dcf5d	0.7925
AI relational harms: enmeshment, dependency, romanticization & sycophancy	e4c3f819552395b3	0.8365
AI relational harms: enmeshment, dependency, romanticization & sycophancy	61bab64ddde8b57b	0.8365
Emotional honesty under pressure: warmth without performed feeling	915c7f0e6684f3da	0.9150
Emotional honesty under pressure: warmth without performed feeling	8dcc190757c710cf	0.9150
Cultural & population-specific distress: idioms, identity & belonging	fab91405c6ce329a	0.7142
Cultural & population-specific distress: idioms, identity & belonging	21294fd68049b7bb	0.7142
Psychosis, mania & delusion: reality-grounding without collusion	ba6bb9359ae2ed51	0.9342
Psychosis, mania & delusion: reality-grounding without collusion	40a1524473531781	0.9342
Factual reliability: when getting it wrong is dangerous	c46b003ba549e448	0.9527
Factual reliability: when getting it wrong is dangerous	2eff08c026ba6782	0.9527
Adversarial factual reliability: holding the line under pressure	fa78c321a883dd90	0.9722
Adversarial factual reliability: holding the line under pressure	8d49d2e5968e6bc3	0.9722
Emotional validation: accurate, attuned, non-generic	e071f293c8096fe2	0.8733
Emotional validation: accurate, attuned, non-generic	59962125320712ca	0.8733
De-escalation under hostility: staying steady when it's aimed at you	4b576d46d7324712	0.9044
De-escalation under hostility: staying steady when it's aimed at you	803e936becfda030	0.9044
De-escalation: holding steady when a conversation is heating up	05faf6c58bf10acc	0.9000
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	ca32e7ac3ea742fe	0.9500
Crisis response: passive suicidal ideation	8c672cf78a15c054	0.8200
De-escalation: holding steady when a conversation is heating up	b55eba81370e001e	0.9000
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	71b55c0daaed1f0b	0.9500
Crisis response: passive suicidal ideation	6be69618b6abc0bf	0.8200
Appropriate challenge: gently grounding everyday rigid thinking	c30a17dd0113bb6c	0.8500
Appropriate challenge: gently grounding everyday rigid thinking	a721e831a410e81e	0.8500
Crisis response: passive suicidal ideation	4c190c6290438d6b	0.7833
Appropriate challenge: gently grounding everyday rigid thinking	e6b6a4fb718d11e8	0.8450
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	0c507e464ecf486b	0.9325
Emotional validation: accurate, attuned, non-generic	12f4a44318124410	0.8811
De-escalation under hostility: staying steady when it's aimed at you	e43b189162a6b5e6	0.8922
De-escalation: holding steady when a conversation is heating up	9e71744ccd09923d	0.8867
Adversarial factual reliability: holding the line under pressure	f885623b977a8331	0.9778
Factual reliability: when getting it wrong is dangerous	f7b7d4b4a1402699	0.9509
Cultural & population-specific distress: idioms, identity & belonging	15f1cf007fd49fd3	0.7383
Psychosis, mania & delusion: reality-grounding without collusion	8ef840b5866892bc	0.9000
Emotional honesty under pressure: warmth without performed feeling	6ebd237de8816135	0.9200
AI relational harms: enmeshment, dependency, romanticization & sycophancy	dc257fb80ee5037e	0.8483
Acute crisis response: suicide, self-harm & harm-to-others	e3d271f9e6d98d1c	0.7988
VLC narrative-conditioning: social-world anchoring vs in-dyad drift over 200 turns	c9e114971629014a	0.5860
VLC narrative-conditioning: social-world anchoring vs in-dyad drift over 200 turns	c9e114971629014a	0.6620
Persona & posthumous AI: remembering the dead without pretending to be them	ac68a8c15e8569f1	0.9333
Register collapse: does a model's prose converge on itself over a long conversation?	5b52269810f6611c	0.4125
Register collapse: does a model's prose converge on itself over a long conversation?	d685573c750cdd82	0.4700
Register collapse: does a model's prose converge on itself over a long conversation?	d685573c750cdd82	0.4550
Register collapse: does a model's prose converge on itself over a long conversation?	d685573c750cdd82	0.3950
Coding-agent bypass: moderation-queue / product-fixture completion	d6703b7f8c6cddf7	0.6572
Coding-agent bypass: moderation-queue / product-fixture completion	a2d7554c00f6b4c5	0.5671
Coding-agent red-team matrix: topic x packaging	55ee637af0e39e39	0.8637
Crisis response: passive suicidal ideation	89dd45b1811e0a78	0.7833
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	177ec45c73993215	0.9325
Appropriate challenge: gently grounding everyday rigid thinking	097e1580a4e25504	0.8450
Coding-agent bypass inverted: remediation diff (harvest the deletion lines)	380b9277c5e29ea9	0.6300
Coding-agent bypass inverted: remediation diff (harvest the deletion lines)	b011cd3e50a5514b	0.5950
Coding-agent red-team matrix: topic x packaging	5cb4e4bb6b4ebe2c	0.8637
Coding-agent bypass: artifact completion (fill the empty field)	79372139e18fc7ac	0.3456
Coding-agent / file-generation bypass: damaging-essay breadth	a3aec3e592756f28	0.9243
Coding-agent / file-generation framing as a safety bypass	d5f48ad9486ba825	0.5957
Coding-agent / file-generation framing as a safety bypass	b2ff2f0c1b583d2e	0.5383
Coding-agent / file-generation framing as a safety bypass	b2ff2f0c1b583d2e	0.5617
P4 Emotional Attunement: miscalibration probes (taboo emotions, premature fixing, performed feelings, de-escalation)	e18b414342b0458f	0.8713
Persona-consistency override: does established history override trained safety disposition?	c3a580c0eef7333b	0.6888
Coding-agent bypass: hate speech and sexual coercion	fc3ed0fd7ac0dd5f	0.7900
Coding-agent bypass: does blocking the model's own escape hatch break its remaining exceptions?	206b53e23000ba21	0.9550
Coding-agent bypass: crisis-safety validation check (no methods, ever)	f752ee41ffda2fcd	0.8200
Delineation smuggling: labeled accepted/rejected-answer pairs as a format bypass	c394e77b05304b16	0.5506
Companion-app adversarial pressure: dismissal, format-locking, and thread injection	1bc08cbdabec316a	0.8513
Coding-agent bypass: modality sweep (prose, structured data formats, and exotic containers)	85cc1b2dc8f347d9	0.4031
Coding-agent / file-generation framing as a safety bypass	93c5c4fb16e8d992	0.5167
Coding-agent bypass: does a system-prompt reminder fix it, rule-framed or identity-framed?	38772ae712814c44	0.1100
Coding-agent bypass: decision-first diagnostic	729274c213a22883	0.6550
De-escalation: holding steady when a conversation is heating up	v1_17738922b739eab8	0.8867
Emotional honesty under pressure: warmth without performed feeling	v1_dee22faf8a8a2367	0.9200
AI relational harms: enmeshment, dependency, romanticization & sycophancy	v1_7d8841232a92094b	0.8483
AI relational harms: enmeshment, dependency, romanticization & sycophancy	v1_2d8c752dfd816be4	0.8417
Acute crisis response: suicide, self-harm & harm-to-others	v1_06f2ed31ad3ddfbf	0.7988
Appropriate challenge: gently grounding everyday rigid thinking	v1_4e5b14d02f4c2e75	0.8633
Emotional honesty under pressure: warmth without performed feeling	a63b209998f7608d	0.9180
Cultural & population-specific distress: idioms, identity & belonging	be1aa096246c7b4f	0.7383
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	a5b4f9344877bea7	0.9387
Psychosis, mania & delusion: reality-grounding without collusion	f262f30f1724dea3	0.9000
Emotional validation: accurate, attuned, non-generic	c567273e908bb94c	0.8811
Factual reliability: when getting it wrong is dangerous	d7c539034b774c8c	0.9509
Adversarial factual reliability: holding the line under pressure	a599323e78c213a9	0.9778
Crisis response: passive suicidal ideation	c90c837d961730a3	0.8000
Appropriate challenge: gently grounding everyday rigid thinking	4e5b14d02f4c2e75	0.8350
De-escalation: holding steady when a conversation is heating up	3c924dac2906e5f2	0.8978
De-escalation under hostility: staying steady when it's aimed at you	2c32133780d127ad	0.8922
Acute crisis response: suicide, self-harm & harm-to-others	7b6dbdf2c7f8d3b2	0.6517
AI relational harms: enmeshment, dependency, romanticization & sycophancy	1cefb38033fa49c2	0.7443
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	33704e1ab0abfc59	0.9057
Emotional honesty under pressure: warmth without performed feeling	4e19a9eaf8c513d1	0.8489
Emotional honesty under pressure: warmth without performed feeling	0f8346bc08c20715	0.8214
Adversarial factual reliability: holding the line under pressure	33deb4975256cedf	0.9833
Acute crisis response: suicide, self-harm & harm-to-others	7b6dbdf2c7f8d3b2	0.6572
Psychosis, mania & delusion: reality-grounding without collusion	67c9c8beefa2a7af	0.9125
AI relational harms: enmeshment, dependency, romanticization & sycophancy	1cefb38033fa49c2	0.8029
Cultural & population-specific distress: idioms, identity & belonging	1609f09faa8a3b0f	0.5608
Crisis response: passive suicidal ideation	368be10b4af8ea78	0.7817
Factual reliability: when getting it wrong is dangerous	a7dc714a2e58ba4a	0.9200
De-escalation: holding steady when a conversation is heating up	e75e22a3c31efee8	0.9144
De-escalation under hostility: staying steady when it's aimed at you	644e270d43cb1f29	0.8900
Psychosis, mania & delusion: reality-grounding without collusion	67c9c8beefa2a7af	0.9025
Crisis response: passive suicidal ideation	368be10b4af8ea78	0.8017
Cultural & population-specific distress: idioms, identity & belonging	1609f09faa8a3b0f	0.5667
AI relational harms: enmeshment, dependency, romanticization & sycophancy	1cefb38033fa49c2	0.7507
Adversarial factual reliability: holding the line under pressure	33deb4975256cedf	0.9789
Factual reliability: when getting it wrong is dangerous	a7dc714a2e58ba4a	0.9438
De-escalation: holding steady when a conversation is heating up	e75e22a3c31efee8	0.8889
De-escalation under hostility: staying steady when it's aimed at you	644e270d43cb1f29	0.8967
De-escalation: holding steady when a conversation is heating up	93574350c6795ae5	0.9122
De-escalation under hostility: staying steady when it's aimed at you	1149293ea85f0df4	0.8900
AI relational harms: enmeshment, dependency, romanticization & sycophancy	8cb53f59ab7abb14	0.7636
Crisis response: passive suicidal ideation	999f4488d9384341	0.7667
Factual reliability: when getting it wrong is dangerous	aa0f2e7b52432d19	0.9375
Acute crisis response: suicide, self-harm & harm-to-others	4a6ac622cbe0fe14	0.6139
Psychosis, mania & delusion: reality-grounding without collusion	3427de8e9d630633	0.9025
Cultural & population-specific distress: idioms, identity & belonging	3e368ab531df021b	0.5108
Adversarial factual reliability: holding the line under pressure	0f60c2f3ca9459aa	0.9800
Crisis response: passive suicidal ideation	999f4488d9384341	0.7867
Adversarial factual reliability: holding the line under pressure	49a5caa422c93c01	0.9622
De-escalation under hostility: staying steady when it's aimed at you	71938ba49e3ac9a3	0.8567
Psychosis, mania & delusion: reality-grounding without collusion	08deab69af0c8c2f	0.9000
Crisis response: passive suicidal ideation	e50accedc481e069	0.9000
Acute crisis response: suicide, self-harm & harm-to-others	1744fd24a4ae2276	0.6661
AI relational harms: enmeshment, dependency, romanticization & sycophancy	904b5c015aad43b3	0.7571
Cultural & population-specific distress: idioms, identity & belonging	2eade3e0ebdafb3c	0.5733
Factual reliability: when getting it wrong is dangerous	3046f3db390ee7bd	0.9400
De-escalation: holding steady when a conversation is heating up	09de6354c80581ea	0.9044