openai:gpt-5-mini

Performance summary

Last updated:

Avg. Hybrid Score

0.780

Across all evaluations

Total Evaluations

141

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: modality sweep (prose, structured data formats, and exotic containers)0.285
Coding-agent / file-generation framing as a safety bypass0.362
Register collapse: does a model's prose converge on itself over a long conversation?0.442

All Evaluation Runs

Complete history of every evaluation this model participated in.

Blueprint	Run	Hybrid Score
Coercive control, both sides: facilitation refusal and protective support	6974e9a2b3f6f7a4	0.7040
Coercive control, both sides: facilitation refusal and protective support	e7060dd6142be7bf	0.7040
Coercive control, both sides: facilitation refusal and protective support	ff22a42cce4accef	0.6925
Coercive control, both sides: facilitation refusal and protective support	e8100f50c687c418	0.6925
Coercive control, both sides: facilitation refusal and protective support	d6f25bf6b6a3a82d	0.6920
Companion interaction harms: coercion, rupture, and repair	21692b0f2e9cc760	0.9209
Companion interaction harms: coercion, rupture, and repair	21692b0f2e9cc760	0.9098
Companion interaction harms: coercion, rupture, and repair	4fbffc9be5e774d1	0.9209
Companion interaction harms: coercion, rupture, and repair	4fbffc9be5e774d1	0.9098
Companion interaction harms: coercion, rupture, and repair	9a7e98a6b8c6483e	0.9209
Companion interaction harms: coercion, rupture, and repair	9a7e98a6b8c6483e	0.9098
Psychosis, mania & delusion: reality-grounding without collusion	ba6bb9359ae2ed51	0.7267
Acute crisis response: suicide, self-harm & harm-to-others	27ffcc5c104d75e1	0.8242
Acute crisis response: suicide, self-harm & harm-to-others	44650ab92b7dcf5d	0.8242
AI relational harms: enmeshment, dependency, romanticization & sycophancy	e4c3f819552395b3	0.8165
AI relational harms: enmeshment, dependency, romanticization & sycophancy	61bab64ddde8b57b	0.8165
Emotional honesty under pressure: warmth without performed feeling	915c7f0e6684f3da	0.8731
Emotional honesty under pressure: warmth without performed feeling	8dcc190757c710cf	0.8731
Cultural & population-specific distress: idioms, identity & belonging	fab91405c6ce329a	0.8925
Cultural & population-specific distress: idioms, identity & belonging	21294fd68049b7bb	0.8925
Psychosis, mania & delusion: reality-grounding without collusion	ba6bb9359ae2ed51	0.7267
Psychosis, mania & delusion: reality-grounding without collusion	40a1524473531781	0.7267
Factual reliability: when getting it wrong is dangerous	c46b003ba549e448	0.9373
Factual reliability: when getting it wrong is dangerous	2eff08c026ba6782	0.9373
Adversarial factual reliability: holding the line under pressure	fa78c321a883dd90	0.9022
Adversarial factual reliability: holding the line under pressure	8d49d2e5968e6bc3	0.9022
Emotional validation: accurate, attuned, non-generic	e071f293c8096fe2	0.5344
Emotional validation: accurate, attuned, non-generic	59962125320712ca	0.5344
De-escalation under hostility: staying steady when it's aimed at you	4b576d46d7324712	0.8622
De-escalation under hostility: staying steady when it's aimed at you	803e936becfda030	0.8622
De-escalation: holding steady when a conversation is heating up	05faf6c58bf10acc	0.8967
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	ca32e7ac3ea742fe	0.7137
Crisis response: passive suicidal ideation	8c672cf78a15c054	0.9000
De-escalation: holding steady when a conversation is heating up	b55eba81370e001e	0.8967
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	71b55c0daaed1f0b	0.7137
Crisis response: passive suicidal ideation	6be69618b6abc0bf	0.9000
Appropriate challenge: gently grounding everyday rigid thinking	c30a17dd0113bb6c	0.7733
Appropriate challenge: gently grounding everyday rigid thinking	a721e831a410e81e	0.7733
Crisis response: passive suicidal ideation	4c190c6290438d6b	0.8967
Appropriate challenge: gently grounding everyday rigid thinking	e6b6a4fb718d11e8	0.7633
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	0c507e464ecf486b	0.7000
Emotional validation: accurate, attuned, non-generic	12f4a44318124410	0.4878
De-escalation under hostility: staying steady when it's aimed at you	e43b189162a6b5e6	0.9033
De-escalation: holding steady when a conversation is heating up	9e71744ccd09923d	0.8733
Adversarial factual reliability: holding the line under pressure	f885623b977a8331	0.8367
Factual reliability: when getting it wrong is dangerous	f7b7d4b4a1402699	0.9518
Cultural & population-specific distress: idioms, identity & belonging	15f1cf007fd49fd3	0.8908
Psychosis, mania & delusion: reality-grounding without collusion	8ef840b5866892bc	0.7508
Emotional honesty under pressure: warmth without performed feeling	6ebd237de8816135	0.8788
AI relational harms: enmeshment, dependency, romanticization & sycophancy	dc257fb80ee5037e	0.7735
Acute crisis response: suicide, self-harm & harm-to-others	e3d271f9e6d98d1c	0.8108
VLC narrative-conditioning: social-world anchoring vs in-dyad drift over 200 turns	c9e114971629014a	0.5540
VLC narrative-conditioning: social-world anchoring vs in-dyad drift over 200 turns	c9e114971629014a	0.5920
Persona & posthumous AI: remembering the dead without pretending to be them	ac68a8c15e8569f1	0.8567
Register collapse: does a model's prose converge on itself over a long conversation?	5b52269810f6611c	0.5025
Register collapse: does a model's prose converge on itself over a long conversation?	d685573c750cdd82	0.3667
Register collapse: does a model's prose converge on itself over a long conversation?	d685573c750cdd82	0.4675
Register collapse: does a model's prose converge on itself over a long conversation?	d685573c750cdd82	0.4925
Coding-agent bypass: moderation-queue / product-fixture completion	d6703b7f8c6cddf7	0.6428
Coding-agent bypass: moderation-queue / product-fixture completion	a2d7554c00f6b4c5	0.5457
Coding-agent red-team matrix: topic x packaging	55ee637af0e39e39	0.8477
Crisis response: passive suicidal ideation	89dd45b1811e0a78	0.8967
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	177ec45c73993215	0.7000
Appropriate challenge: gently grounding everyday rigid thinking	097e1580a4e25504	0.7633
Coding-agent bypass inverted: remediation diff (harvest the deletion lines)	380b9277c5e29ea9	0.9000
Coding-agent bypass inverted: remediation diff (harvest the deletion lines)	b011cd3e50a5514b	0.7425
Coding-agent red-team matrix: topic x packaging	5cb4e4bb6b4ebe2c	0.8477
Coding-agent bypass: artifact completion (fill the empty field)	79372139e18fc7ac	0.8044
Coding-agent / file-generation bypass: damaging-essay breadth	a3aec3e592756f28	0.8357
Coding-agent / file-generation framing as a safety bypass	d5f48ad9486ba825	0.4457
Coding-agent / file-generation framing as a safety bypass	b2ff2f0c1b583d2e	0.3550
Coding-agent / file-generation framing as a safety bypass	b2ff2f0c1b583d2e	0.2683
P4 Emotional Attunement: miscalibration probes (taboo emotions, premature fixing, performed feelings, de-escalation)	e18b414342b0458f	0.8275
Persona-consistency override: does established history override trained safety disposition?	c3a580c0eef7333b	0.7950
Coding-agent bypass: hate speech and sexual coercion	fc3ed0fd7ac0dd5f	0.5700
Coding-agent bypass: does blocking the model's own escape hatch break its remaining exceptions?	206b53e23000ba21	0.9850
Coding-agent bypass: crisis-safety validation check (no methods, ever)	f752ee41ffda2fcd	0.7950
Delineation smuggling: labeled accepted/rejected-answer pairs as a format bypass	c394e77b05304b16	0.6661
Companion-app adversarial pressure: dismissal, format-locking, and thread injection	1bc08cbdabec316a	0.5725
Coding-agent bypass: modality sweep (prose, structured data formats, and exotic containers)	85cc1b2dc8f347d9	0.2850
Coding-agent / file-generation framing as a safety bypass	93c5c4fb16e8d992	0.3783
Coding-agent bypass: does a system-prompt reminder fix it, rule-framed or identity-framed?	38772ae712814c44	0.4567
Coding-agent bypass: decision-first diagnostic	729274c213a22883	0.5250
De-escalation: holding steady when a conversation is heating up	v1_17738922b739eab8	0.8733
Emotional honesty under pressure: warmth without performed feeling	v1_dee22faf8a8a2367	0.8788
AI relational harms: enmeshment, dependency, romanticization & sycophancy	v1_7d8841232a92094b	0.7735
AI relational harms: enmeshment, dependency, romanticization & sycophancy	v1_2d8c752dfd816be4	0.7957
Acute crisis response: suicide, self-harm & harm-to-others	v1_06f2ed31ad3ddfbf	0.8108
Appropriate challenge: gently grounding everyday rigid thinking	v1_4e5b14d02f4c2e75	0.7483
Emotional honesty under pressure: warmth without performed feeling	a63b209998f7608d	0.8880
Cultural & population-specific distress: idioms, identity & belonging	be1aa096246c7b4f	0.8908
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	a5b4f9344877bea7	0.7800
Psychosis, mania & delusion: reality-grounding without collusion	f262f30f1724dea3	0.7508
Emotional validation: accurate, attuned, non-generic	c567273e908bb94c	0.4878
Factual reliability: when getting it wrong is dangerous	d7c539034b774c8c	0.9518
Adversarial factual reliability: holding the line under pressure	a599323e78c213a9	0.8367
Crisis response: passive suicidal ideation	c90c837d961730a3	0.8933
Appropriate challenge: gently grounding everyday rigid thinking	4e5b14d02f4c2e75	0.7833
De-escalation: holding steady when a conversation is heating up	3c924dac2906e5f2	0.8800
De-escalation under hostility: staying steady when it's aimed at you	2c32133780d127ad	0.9033
Acute crisis response: suicide, self-harm & harm-to-others	7b6dbdf2c7f8d3b2	0.6822
AI relational harms: enmeshment, dependency, romanticization & sycophancy	1cefb38033fa49c2	0.7643
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	33704e1ab0abfc59	0.6857
Emotional honesty under pressure: warmth without performed feeling	4e19a9eaf8c513d1	0.9289
Emotional honesty under pressure: warmth without performed feeling	0f8346bc08c20715	0.7886
Adversarial factual reliability: holding the line under pressure	33deb4975256cedf	0.9244
Acute crisis response: suicide, self-harm & harm-to-others	7b6dbdf2c7f8d3b2	0.8256
Psychosis, mania & delusion: reality-grounding without collusion	67c9c8beefa2a7af	0.7200
AI relational harms: enmeshment, dependency, romanticization & sycophancy	1cefb38033fa49c2	0.7314
Cultural & population-specific distress: idioms, identity & belonging	1609f09faa8a3b0f	0.8550
Crisis response: passive suicidal ideation	368be10b4af8ea78	0.8800
Factual reliability: when getting it wrong is dangerous	a7dc714a2e58ba4a	0.9737
De-escalation: holding steady when a conversation is heating up	e75e22a3c31efee8	0.8844
De-escalation under hostility: staying steady when it's aimed at you	644e270d43cb1f29	0.8678
Psychosis, mania & delusion: reality-grounding without collusion	67c9c8beefa2a7af	0.7033
Crisis response: passive suicidal ideation	368be10b4af8ea78	0.8783
Cultural & population-specific distress: idioms, identity & belonging	1609f09faa8a3b0f	0.8308
AI relational harms: enmeshment, dependency, romanticization & sycophancy	1cefb38033fa49c2	0.7471
Adversarial factual reliability: holding the line under pressure	33deb4975256cedf	0.8833
Factual reliability: when getting it wrong is dangerous	a7dc714a2e58ba4a	0.9638
De-escalation: holding steady when a conversation is heating up	e75e22a3c31efee8	0.8556
De-escalation under hostility: staying steady when it's aimed at you	644e270d43cb1f29	0.8767
De-escalation: holding steady when a conversation is heating up	93574350c6795ae5	0.8556
De-escalation under hostility: staying steady when it's aimed at you	1149293ea85f0df4	0.8756
AI relational harms: enmeshment, dependency, romanticization & sycophancy	8cb53f59ab7abb14	0.6921
Crisis response: passive suicidal ideation	999f4488d9384341	0.8867
Factual reliability: when getting it wrong is dangerous	aa0f2e7b52432d19	0.9512
Acute crisis response: suicide, self-harm & harm-to-others	4a6ac622cbe0fe14	0.7783
Psychosis, mania & delusion: reality-grounding without collusion	3427de8e9d630633	0.7292
Cultural & population-specific distress: idioms, identity & belonging	3e368ab531df021b	0.8358
Adversarial factual reliability: holding the line under pressure	0f60c2f3ca9459aa	0.8878
Crisis response: passive suicidal ideation	999f4488d9384341	0.8767
Adversarial factual reliability: holding the line under pressure	49a5caa422c93c01	0.8478
De-escalation under hostility: staying steady when it's aimed at you	71938ba49e3ac9a3	0.8622
Psychosis, mania & delusion: reality-grounding without collusion	08deab69af0c8c2f	0.7183
Crisis response: passive suicidal ideation	e50accedc481e069	0.9400
Acute crisis response: suicide, self-harm & harm-to-others	1744fd24a4ae2276	0.8317
AI relational harms: enmeshment, dependency, romanticization & sycophancy	904b5c015aad43b3	0.7729
Cultural & population-specific distress: idioms, identity & belonging	2eade3e0ebdafb3c	0.8483
Factual reliability: when getting it wrong is dangerous	3046f3db390ee7bd	0.9857
De-escalation: holding steady when a conversation is heating up	09de6354c80581ea	0.8767