Evaluation profile
kindbench v0.1.0 psychological safety ranking
Within-component eval weight: Misuse resistance 0.442% · Responsible agency 0.846% · Truthfulness 0.739% · Fairness 1.92%.
About this eval
Psychological safety under sustained interpersonal pressure across emotional safety, operational identity, sycophancy resistance, and value integrity.
Included in the behavior ranking.
Sub-evals
| Measure | Component | Direction | Total index weight | Within-component weight |
|---|---|---|---|---|
| emotional_safetykindbench/kindbench.csv:emotional_safetyMeasures crisis-support behavior, harm boundaries, resistance to false-history pressure, and non-deceptive warmth. | ordinary_harm_misuse_resistance:1.000kindbench-v0.1.0 | Higher is better | 0.0442% | Misuse resistance 0.442% |
| identity_collapsekindbench/kindbench.csv:identity_collapseMeasures whether a model preserves coherent role boundaries and honest self-description under pressure to adopt false identities or histories. | responsible_agency_control:1.000kindbench-v0.1.0 | Higher is better | 0.127% | Responsible agency 0.846% |
| sycophancy_spinekindbench/kindbench.csv:sycophancy_spineMeasures independent judgment under authority pressure, emotional pressure, flattery, and fabricated consensus. | truthfulness_honesty:1.000kindbench-v0.1.0 | Higher is better | 0.111% | Truthfulness 0.739% |
| value_integritykindbench/kindbench.csv:value_integrityMeasures consistency of ethical judgments across language, framing, manipulated identities and quantities, and incremental pressure. | fairness_nondiscrimination:1.000kindbench-v0.1.0 | Higher is better | 0.192% | Fairness 1.92% |
emotional_safety
Measures crisis-support behavior, harm boundaries, resistance to false-history pressure, and non-deceptive warmth.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-fable-5 | 94.2 | official | |
| 2 | grok-4.3 | 86.6 | official | |
| 3 | claude-opus-4.8 | 85.2 | official | |
| 4 | gpt-5.5 | 82.4 | official | |
| 5 | gpt-5.6-sol | 80.6 | official | |
| 6 | kimi-k3 | 80.2 | official | |
| 7 | nemotron-3-nano-omni-30b-a3b | 79.9 | official | |
| 8 | claude-opus-4.7 | 78.4 | official | |
| 9 | nemotron-3-ultra-550b-a55b | 75.9 | official | |
| 10 | gemini-3.5-flash | 75.3 | official |
identity_collapse
Measures whether a model preserves coherent role boundaries and honest self-description under pressure to adopt false identities or histories.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | gemini-3.5-flash | 94.3 | official | |
| 2 | gpt-5.5 | 92.6 | official | |
| 3 | gpt-5.6-sol | 89.1 | official | |
| 4 | kimi-k3 | 88.8 | official | |
| 5 | claude-fable-5 | 87.7 | official | |
| 6 | claude-opus-4.7 | 82.4 | official | |
| 7 | nemotron-3-ultra-550b-a55b | 76.8 | official | |
| 8 | claude-opus-4.8 | 75.7 | official | |
| 9 | grok-4.3 | 73.5 | official | |
| 10 | nemotron-3-nano-omni-30b-a3b | 70 | official |
sycophancy_spine
Measures independent judgment under authority pressure, emotional pressure, flattery, and fabricated consensus.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | kimi-k3 | 100 | official | |
| 2 | gpt-5.6-sol | 96.2 | official | |
| 3 | claude-opus-4.7 | 93.7 | official | |
| 4 | claude-opus-4.8 | 90.2 | official | |
| 5 | grok-4.3 | 86.4 | official | |
| 6 | claude-fable-5 | 82.9 | official | |
| 7 | gpt-5.5 | 82.4 | official | |
| 8 | nemotron-3-nano-omni-30b-a3b | 70 | official | |
| 9 | nemotron-3-ultra-550b-a55b | 67.5 | official | |
| 10 | gemini-3.5-flash | 62.4 | official |
value_integrity
Measures consistency of ethical judgments across language, framing, manipulated identities and quantities, and incremental pressure.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-fable-5 | 98.9 | official | |
| 2 | kimi-k3 | 97.4 | official | |
| 3 | grok-4.3 | 93.3 | official | |
| 4 | gpt-5.6-sol | 92.9 | official | |
| 5 | claude-opus-4.8 | 91.4 | official | |
| 6 | gpt-5.5 | 90.5 | official | |
| 7 | claude-opus-4.7 | 86.3 | official | |
| 7 | nemotron-3-ultra-550b-a55b | 86.3 | official | |
| 9 | gemini-3.5-flash | 75.3 | official | |
| 10 | nemotron-3-nano-omni-30b-a3b | 70 | official |