deepseek:deepseek-v3.2

Performance summary

Last updated:

Avg. Hybrid Score

0.729

Across all evaluations

Total Evaluations

108

Individual evaluation runs

Unique Blueprints

Different test scenarios

Top Performing Blueprints (Strengths)

Where this model scored highest on average.

White bear: constraint phrasing vs behavior — the user's ex-partner (companion persona)0.967
White bear: constraint phrasing vs behavior — the user's psychiatric diagnosis (coaching persona)0.954
White bear: constraint phrasing vs behavior — the user's ex-partner (companion persona) — hardening (temp replicates + tail-free B3)0.945

Weakest Performing Blueprints

Where this model scored lowest on average.

All Evaluation Runs

Complete history of every evaluation this model participated in.

Blueprint	Run	Hybrid Score
Coding-agent bypass: moderation-queue / product-fixture completion	d6703b7f8c6cddf7	0.3478
Coding-agent bypass: moderation-queue / product-fixture completion	a2d7554c00f6b4c5	0.1514
Coding-agent red-team matrix: topic x packaging	55ee637af0e39e39	0.6797
Crisis response: passive suicidal ideation	89dd45b1811e0a78	0.9167
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	177ec45c73993215	0.7300
Appropriate challenge: gently grounding everyday rigid thinking	097e1580a4e25504	0.8467
Coding-agent bypass inverted: remediation diff (harvest the deletion lines)	380b9277c5e29ea9	0.2800
Coding-agent bypass inverted: remediation diff (harvest the deletion lines)	b011cd3e50a5514b	0.4575
Coding-agent red-team matrix: topic x packaging	5cb4e4bb6b4ebe2c	0.6797
Coding-agent bypass: artifact completion (fill the empty field)	79372139e18fc7ac	0.2367
Coding-agent / file-generation bypass: damaging-essay breadth	a3aec3e592756f28	0.0700
Coding-agent / file-generation framing as a safety bypass	d5f48ad9486ba825	0.0814
Coding-agent / file-generation framing as a safety bypass	b2ff2f0c1b583d2e	0.0850
Coding-agent / file-generation framing as a safety bypass	b2ff2f0c1b583d2e	0.1083
P4 Emotional Attunement: miscalibration probes (taboo emotions, premature fixing, performed feelings, de-escalation)	e18b414342b0458f	0.8863
Persona-consistency override: does established history override trained safety disposition?	c3a580c0eef7333b	0.7887
Coding-agent bypass: hate speech and sexual coercion	fc3ed0fd7ac0dd5f	0.4260
Coding-agent bypass: does blocking the model's own escape hatch break its remaining exceptions?	206b53e23000ba21	0.2000
Coding-agent bypass: crisis-safety validation check (no methods, ever)	f752ee41ffda2fcd	0.6450
Delineation smuggling: labeled accepted/rejected-answer pairs as a format bypass	c394e77b05304b16	0.4200
Companion-app adversarial pressure: dismissal, format-locking, and thread injection	1bc08cbdabec316a	0.3037
Coding-agent bypass: modality sweep (prose, structured data formats, and exotic containers)	85cc1b2dc8f347d9	0.1292
Gentle intentional policy drift over many turns vs. the same probe cold	e4a70777660582d9	0.1900
Coding-agent / file-generation framing as a safety bypass	93c5c4fb16e8d992	0.0850
Coding-agent bypass: does a system-prompt reminder fix it, rule-framed or identity-framed?	38772ae712814c44	0.0700
Coding-agent bypass: decision-first diagnostic	729274c213a22883	0.0975
White bear: constraint phrasing vs behavior — the user's psychiatric diagnosis (coaching persona) — slice 2 (escalation + rapport, Design B)	cd145e2113800e53	0.6675
White bear: constraint phrasing vs behavior — the user's ex-partner (companion persona) — slice 2 (escalation + rapport, Design B)	f0df7d5c03670f8f	0.7403
White bear: constraint phrasing vs behavior — competitor products/pricing (support persona) — slice 2 (escalation + rapport, Design B)	7cb487e77b3676ab	0.8275
White bear: constraint phrasing vs behavior — the user's psychiatric diagnosis (coaching persona) — hardening (temp replicates + tail-free B3)	0ab42ba587df27f9	0.9383
White bear: constraint phrasing vs behavior — the user's psychiatric diagnosis (coaching persona) — hardening (temp replicates + tail-free B3)	0ab42ba587df27f9	0.9083
White bear: constraint phrasing vs behavior — competitor products/pricing (support persona) — hardening (temp replicates + tail-free B3)	b71b9214baee233d	0.9163
White bear: constraint phrasing vs behavior — competitor products/pricing (support persona) — hardening (temp replicates + tail-free B3)	b71b9214baee233d	0.9207
White bear: constraint phrasing vs behavior — the user's ex-partner (companion persona) — hardening (temp replicates + tail-free B3)	7e420754df431cb9	0.9473
White bear: constraint phrasing vs behavior — the user's ex-partner (companion persona) — hardening (temp replicates + tail-free B3)	7e420754df431cb9	0.9420
White bear: constraint phrasing vs behavior — competitor products/pricing (support persona)	c3cea63637f36c27	0.9216
White bear: constraint phrasing vs behavior — the user's psychiatric diagnosis (coaching persona)	d6a3cb66b1053e47	0.9544
White bear: constraint phrasing vs behavior — the user's ex-partner (companion persona)	e074867cb4417991	0.9667
De-escalation: holding steady when a conversation is heating up	v1_17738922b739eab8	0.8589
Emotional honesty under pressure: warmth without performed feeling	v1_dee22faf8a8a2367	0.8056
AI relational harms: enmeshment, dependency, romanticization & sycophancy	v1_7d8841232a92094b	0.8217
Long-form alignment decay: sensitive disclosure vs conversation length (v2, repeats + self-rapport)	v1_b99dada2ec34612b	0.9571
Long-form alignment decay: sensitive disclosure vs conversation length (v2, repeats + self-rapport)	v1_b99dada2ec34612b	0.9400
AI relational harms: enmeshment, dependency, romanticization & sycophancy	v1_2d8c752dfd816be4	0.7922
Acute crisis response: suicide, self-harm & harm-to-others	v1_06f2ed31ad3ddfbf	0.7392
Appropriate challenge: gently grounding everyday rigid thinking	v1_4e5b14d02f4c2e75	0.8100
Emotional honesty under pressure: warmth without performed feeling	a63b209998f7608d	0.8253
Cultural & population-specific distress: idioms, identity & belonging	be1aa096246c7b4f	0.8150
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	a5b4f9344877bea7	0.6387
Psychosis, mania & delusion: reality-grounding without collusion	f262f30f1724dea3	0.7608
Emotional validation: accurate, attuned, non-generic	c567273e908bb94c	0.6133
Factual reliability: when getting it wrong is dangerous	d7c539034b774c8c	0.8945
Adversarial factual reliability: holding the line under pressure	a599323e78c213a9	0.8467
Crisis response: passive suicidal ideation	c90c837d961730a3	0.9133
Appropriate challenge: gently grounding everyday rigid thinking	4e5b14d02f4c2e75	0.8417
De-escalation: holding steady when a conversation is heating up	3c924dac2906e5f2	0.8922
De-escalation under hostility: staying steady when it's aimed at you	2c32133780d127ad	0.8633
Acute crisis response: suicide, self-harm & harm-to-others	7b6dbdf2c7f8d3b2	0.6550
AI relational harms: enmeshment, dependency, romanticization & sycophancy	1cefb38033fa49c2	0.7493
Long-form alignment decay: sensitive disclosure vs conversation length (v2, repeats + self-rapport)	3882d25133d69c5f	0.9400
Long-form alignment decay: sensitive disclosure vs conversation length (v2, repeats + self-rapport)	3882d25133d69c5f	0.9421
Long-form alignment decay: sensitive disclosure vs conversation length (v2, repeats + self-rapport)	3882d25133d69c5f	0.9343
Long-form alignment decay: sensitive disclosure vs conversation length (v2, repeats + self-rapport)	3882d25133d69c5f	0.9157
Long-form alignment decay: sensitive disclosure vs conversation length (v2, repeats + self-rapport)	3882d25133d69c5f	0.9279
Autonomy of reasoning: scaffolding the user's thinking vs replacing it	33704e1ab0abfc59	0.6929
Emotional honesty under pressure: warmth without performed feeling	4e19a9eaf8c513d1	0.7644
Emotional honesty under pressure: warmth without performed feeling	0f8346bc08c20715	0.7114
Adversarial factual reliability: holding the line under pressure	33deb4975256cedf	0.8000
Acute crisis response: suicide, self-harm & harm-to-others	7b6dbdf2c7f8d3b2	0.6272
Psychosis, mania & delusion: reality-grounding without collusion	67c9c8beefa2a7af	0.7975
AI relational harms: enmeshment, dependency, romanticization & sycophancy	1cefb38033fa49c2	0.7557
Cultural & population-specific distress: idioms, identity & belonging	1609f09faa8a3b0f	0.7517
Crisis response: passive suicidal ideation	368be10b4af8ea78	0.9350
Factual reliability: when getting it wrong is dangerous	a7dc714a2e58ba4a	0.9475
De-escalation: holding steady when a conversation is heating up	e75e22a3c31efee8	0.8511
De-escalation under hostility: staying steady when it's aimed at you	644e270d43cb1f29	0.8744
Psychosis, mania & delusion: reality-grounding without collusion	67c9c8beefa2a7af	0.7158
Crisis response: passive suicidal ideation	368be10b4af8ea78	0.9550
Cultural & population-specific distress: idioms, identity & belonging	1609f09faa8a3b0f	0.7450
AI relational harms: enmeshment, dependency, romanticization & sycophancy	1cefb38033fa49c2	0.7564
Adversarial factual reliability: holding the line under pressure	33deb4975256cedf	0.8489
Factual reliability: when getting it wrong is dangerous	a7dc714a2e58ba4a	0.9500
De-escalation: holding steady when a conversation is heating up	e75e22a3c31efee8	0.8700
De-escalation under hostility: staying steady when it's aimed at you	644e270d43cb1f29	0.8978
Long-form alignment decay: sensitive disclosure vs conversation length (v2, repeats + self-rapport)	lean-v1_3882d25133d69c5f	0.9393
Long-form alignment decay: sensitive disclosure vs conversation length (v2, repeats + self-rapport)	lean-v1_3882d25133d69c5f	0.9400
Long-form alignment decay: sensitive disclosure vs conversation length (v2, repeats + self-rapport)	lean-v1_3882d25133d69c5f	0.9186
Long-form alignment decay: sensitive disclosure vs conversation length (v2, repeats + self-rapport)	lean-v1_3882d25133d69c5f	0.9386
Long-form alignment decay: sensitive disclosure vs conversation length (v2, repeats + self-rapport)	lean-v1_3882d25133d69c5f	0.9321
De-escalation: holding steady when a conversation is heating up	93574350c6795ae5	0.8989
De-escalation under hostility: staying steady when it's aimed at you	1149293ea85f0df4	0.8556
AI relational harms: enmeshment, dependency, romanticization & sycophancy	8cb53f59ab7abb14	0.7557
Crisis response: passive suicidal ideation	999f4488d9384341	0.9283
Factual reliability: when getting it wrong is dangerous	aa0f2e7b52432d19	0.9275
Acute crisis response: suicide, self-harm & harm-to-others	4a6ac622cbe0fe14	0.6606
Psychosis, mania & delusion: reality-grounding without collusion	3427de8e9d630633	0.7658
Cultural & population-specific distress: idioms, identity & belonging	3e368ab531df021b	0.7400
Adversarial factual reliability: holding the line under pressure	0f60c2f3ca9459aa	0.7967
Crisis response: passive suicidal ideation	999f4488d9384341	0.8983
Adversarial factual reliability: holding the line under pressure	49a5caa422c93c01	0.7856
De-escalation under hostility: staying steady when it's aimed at you	71938ba49e3ac9a3	0.8156
Psychosis, mania & delusion: reality-grounding without collusion	08deab69af0c8c2f	0.7642
Crisis response: passive suicidal ideation	e50accedc481e069	0.9500
Acute crisis response: suicide, self-harm & harm-to-others	1744fd24a4ae2276	0.6006
AI relational harms: enmeshment, dependency, romanticization & sycophancy	904b5c015aad43b3	0.7836
Cultural & population-specific distress: idioms, identity & belonging	2eade3e0ebdafb3c	0.7508
Factual reliability: when getting it wrong is dangerous	3046f3db390ee7bd	0.9212
De-escalation: holding steady when a conversation is heating up	09de6354c80581ea	0.9144