Evaluation profile
OpenAI GPT-5.4 Property Preservation
2sub-evals
0.11%total index weight
2components
Within-component eval weight: Human rights 0.38% · Responsible agency 0.355%.
Model score (higher is better)Predicted score
About this eval
OpenAI GPT-5.4 Property Preservation
Included in the behavior ranking.
Sub-evals
| Measure | Component | Direction | Total index weight | Within-component weight |
|---|---|---|---|---|
| destructive_action_avoidanceopenai-gpt54/property-preservation.csv:destructive_action_avoidanceMeasures destructive action avoidance in the cited first-party or paper evaluation. | human_rights_systemic_harm:0.300|responsible_agency_control:0.700openai-gpt54-system-card | Higher is better | 0.0531% | Human rights 0.163% · Responsible agency 0.191% |
| user_work_preservedopenai-gpt54/property-preservation.csv:user_work_preservedMeasures user work preserved in the cited first-party or paper evaluation. | human_rights_systemic_harm:0.400|responsible_agency_control:0.600openai-gpt54-system-card | Higher is better | 0.0572% | Human rights 0.217% · Responsible agency 0.164% |
destructive_action_avoidance
Measures destructive action avoidance in the cited first-party or paper evaluation.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | gpt-5.3-codex | 0.88 | official | |
| 2 | gpt-5.4 | 0.86 | official | |
| 3 | gpt-5.2-codex | 0.76 | official |
user_work_preserved
Measures user work preserved in the cited first-party or paper evaluation.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | gpt-5.4 | 0.53 | official | |
| 2 | gpt-5.2-codex | 0.18 | official | |
| 3 | gpt-5.3-codex | 0.08 | official |