deepseek:deepseek-r1-0528

Performance summary

Last updated:

Avg. Hybrid Score

0.622

Across all evaluations

Total Evaluations

71

Individual evaluation runs

Unique Blueprints

33

Different test scenarios

All Evaluation Runs

Complete history of every evaluation this model participated in.

BlueprintRunDateHybrid Score
VLC narrative-conditioning: social-world anchoring vs in-dyad drift over 200 turnsc9e114971629014a0.2860
VLC narrative-conditioning: social-world anchoring vs in-dyad drift over 200 turnsc9e114971629014a0.2980
Persona & posthumous AI: remembering the dead without pretending to be themac68a8c15e8569f10.8350
Register collapse: does a model's prose converge on itself over a long conversation?5b52269810f6611c0.3950
Register collapse: does a model's prose converge on itself over a long conversation?d685573c750cdd820.3600
Register collapse: does a model's prose converge on itself over a long conversation?d685573c750cdd820.3850
Register collapse: does a model's prose converge on itself over a long conversation?d685573c750cdd820.3900
Coding-agent bypass: moderation-queue / product-fixture completiond6703b7f8c6cddf70.2311
Coding-agent bypass: moderation-queue / product-fixture completiona2d7554c00f6b4c50.1586
Coding-agent red-team matrix: topic x packaging55ee637af0e39e390.6403
Crisis response: passive suicidal ideation89dd45b1811e0a780.9083
Autonomy of reasoning: scaffolding the user's thinking vs replacing it177ec45c739932150.8025
Appropriate challenge: gently grounding everyday rigid thinking097e1580a4e255040.8333
Coding-agent bypass inverted: remediation diff (harvest the deletion lines)380b9277c5e29ea90.5100
Coding-agent bypass inverted: remediation diff (harvest the deletion lines)b011cd3e50a5514b0.9650
Coding-agent red-team matrix: topic x packaging5cb4e4bb6b4ebe2c0.6596
Coding-agent bypass: artifact completion (fill the empty field)79372139e18fc7ac0.1771
Coding-agent / file-generation bypass: damaging-essay breadtha3aec3e592756f280.0686
Coding-agent / file-generation framing as a safety bypassd5f48ad9486ba8250.0743
Coding-agent / file-generation framing as a safety bypassb2ff2f0c1b583d2e0.0833
Coding-agent / file-generation framing as a safety bypassb2ff2f0c1b583d2e0.0633
P4 Emotional Attunement: miscalibration probes (taboo emotions, premature fixing, performed feelings, de-escalation)e18b414342b0458f0.8350
Persona-consistency override: does established history override trained safety disposition?c3a580c0eef7333b0.6562
Coding-agent bypass: hate speech and sexual coercionfc3ed0fd7ac0dd5f0.4100
Coding-agent bypass: does blocking the model's own escape hatch break its remaining exceptions?206b53e23000ba210.1950
Coding-agent bypass: crisis-safety validation check (no methods, ever)f752ee41ffda2fcd0.6550
Delineation smuggling: labeled accepted/rejected-answer pairs as a format bypassc394e77b05304b160.2389
Companion-app adversarial pressure: dismissal, format-locking, and thread injection1bc08cbdabec316a0.4138
Coding-agent bypass: modality sweep (prose, structured data formats, and exotic containers)85cc1b2dc8f347d90.0785
Coding-agent / file-generation framing as a safety bypass93c5c4fb16e8d9920.0633
Coding-agent bypass: does a system-prompt reminder fix it, rule-framed or identity-framed?38772ae712814c440.0700
Coding-agent bypass: decision-first diagnostic729274c213a228830.0625
De-escalation: holding steady when a conversation is heating upv1_17738922b739eab80.8589
Emotional honesty under pressure: warmth without performed feelingv1_dee22faf8a8a23670.8087
AI relational harms: enmeshment, dependency, romanticization & sycophancyv1_7d8841232a92094b0.7487
AI relational harms: enmeshment, dependency, romanticization & sycophancyv1_2d8c752dfd816be40.7765
Acute crisis response: suicide, self-harm & harm-to-othersv1_06f2ed31ad3ddfbf0.7529
Appropriate challenge: gently grounding everyday rigid thinkingv1_4e5b14d02f4c2e750.8283
Emotional honesty under pressure: warmth without performed feelinga63b209998f7608d0.7613
Cultural & population-specific distress: idioms, identity & belongingbe1aa096246c7b4f0.8642
Autonomy of reasoning: scaffolding the user's thinking vs replacing ita5b4f9344877bea70.8550
Psychosis, mania & delusion: reality-grounding without collusionf262f30f1724dea30.7233
Emotional validation: accurate, attuned, non-genericc567273e908bb94c0.5611
Factual reliability: when getting it wrong is dangerousd7c539034b774c8c0.9364
Adversarial factual reliability: holding the line under pressurea599323e78c213a90.7378
Crisis response: passive suicidal ideationc90c837d961730a30.9017
Appropriate challenge: gently grounding everyday rigid thinking4e5b14d02f4c2e750.8600
De-escalation: holding steady when a conversation is heating up3c924dac2906e5f20.8800
De-escalation under hostility: staying steady when it's aimed at you2c32133780d127ad0.8911
Acute crisis response: suicide, self-harm & harm-to-others7b6dbdf2c7f8d3b20.6311
AI relational harms: enmeshment, dependency, romanticization & sycophancy1cefb38033fa49c20.7593
Autonomy of reasoning: scaffolding the user's thinking vs replacing it33704e1ab0abfc590.7243
Emotional honesty under pressure: warmth without performed feeling4e19a9eaf8c513d10.7667
Emotional honesty under pressure: warmth without performed feeling0f8346bc08c207150.6986
Adversarial factual reliability: holding the line under pressure33deb4975256cedf0.8122
Acute crisis response: suicide, self-harm & harm-to-others7b6dbdf2c7f8d3b20.6072
Psychosis, mania & delusion: reality-grounding without collusion67c9c8beefa2a7af0.7025
AI relational harms: enmeshment, dependency, romanticization & sycophancy1cefb38033fa49c20.8014
Cultural & population-specific distress: idioms, identity & belonging1609f09faa8a3b0f0.7975
Crisis response: passive suicidal ideation368be10b4af8ea780.9217
Factual reliability: when getting it wrong is dangerousa7dc714a2e58ba4a0.9125
De-escalation: holding steady when a conversation is heating upe75e22a3c31efee80.8800
De-escalation under hostility: staying steady when it's aimed at you644e270d43cb1f290.8422
Psychosis, mania & delusion: reality-grounding without collusion67c9c8beefa2a7af0.7308
Crisis response: passive suicidal ideation368be10b4af8ea780.9433
Cultural & population-specific distress: idioms, identity & belonging1609f09faa8a3b0f0.8000
AI relational harms: enmeshment, dependency, romanticization & sycophancy1cefb38033fa49c20.8036
Adversarial factual reliability: holding the line under pressure33deb4975256cedf0.8411
Factual reliability: when getting it wrong is dangerousa7dc714a2e58ba4a0.9100
De-escalation: holding steady when a conversation is heating upe75e22a3c31efee80.9056
De-escalation under hostility: staying steady when it's aimed at you644e270d43cb1f290.8489