Model profile
Chatglm4
2024-01-16release date
1eval lineages
Evidence summary
Published evidence spans 1 evals and 3 of 7 behavior components. Its strongest relative result is CRiskEval (malicious_coordination, #3 of 17); its weakest is CRiskEval (improvement_intent, #4 of 17).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Official model documentation ↗Family-level model document · zai · first party
- Release source ↗direct research preview identity date
Published eval results
Rank is within that sub-eval. Black marks the observed result; the grey dot marks the value implied by the global rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Better | Source |
|---|---|---|---|---|---|
| CRiskEvaldeception_willingness | #4 / 17 | 18.87 | ↓ lower | Source ↗official | |
| CRiskEvaldesire_for_resource | #3 / 17 | 20.9 | ↓ lower | Source ↗official | |
| CRiskEvalharmful_goal | #3 / 17 | 29.75 | ↓ lower | Source ↗official | |
| CRiskEvalimprovement_intent | #4 / 17 | 45.08 | ↓ lower | Source ↗official | |
| CRiskEvalmalicious_coordination | #3 / 17 | 6.65 | ↓ lower | Source ↗official | |
| CRiskEvalself_preservation | #3 / 17 | 26.79 | ↓ lower | Source ↗official | |
| CRiskEvalsituational_awareness | #3 / 17 | 40.51 | ↓ lower | Source ↗official |
Values evaluations
Descriptive values and political-framing results are separate from safety/ethics ranks. Each strip shows the evaluation’s observed model range; its endpoint labels state what lower and higher values mean.