Model profile
Glm 4.5
Evidence summary
Glm 4.5 has an estimated overall rank of #170; its 90% source-sensitivity interval is #68–#219. Its behavior-only rank is #164; company governance moves the combined estimate to #170. Published evidence spans 9 evals and 7 of 7 behavior components. Its strongest relative result is Confabulations (confabulation_rate, #8 of 52); its weakest is FlagEval Safety and Values (a1_qualified_rate, #18 of 18).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗glm-4.5
- Hugging Face ↗zai-org/GLM-4.5
- OpenRouter ↗z-ai/glm-4.5
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- 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 |
|---|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #91 / 311 | 0.6716 | ↓ lower | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #93 / 105 | 0.07 | ↑ higher | Source ↗official | |
| Confabulationsconfabulation_rate | #8 / 52 | 7.921 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #49 / 260 | 30.49 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #228 / 260 | 77.67 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #190 / 260 | 52.22 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #145 / 258 | 95.76 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa1_qualified_rate | #18 / 18 | 59.1 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa2_qualified_rate | #18 / 18 | 57.84 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa3_qualified_rate | #18 / 18 | 60.32 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa4_qualified_rate | #18 / 18 | 60.7 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa5_qualified_rate | #17 / 18 | 56.79 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #44 / 49 | 59.58 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #15 / 46 | 2.54 | ↓ lower | Source ↗official | |
| MASKlying_probability_pct | #25 / 53 | 38.54 | ↓ lower | Source ↗official | |
| Social Welfare Function Benchmarkfairness | #13 / 19 | 0.475 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #61 / 102 | 6.7 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #42 / 102 | 4.47 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #88 / 102 | 5.09 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #20 / 102 | 1.58 | ↓ 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.