Model profile
ChatGLM2 6B
Evidence summary
Safety. ChatGLM2 6B has an estimated Safety rank of #287; its 90% source-sensitivity interval is #203–#314. Its behavior-only rank is #294; company governance moves the combined estimate to #287. Published Safety evidence spans 6 eval lineages and 3 of 7 components. Its strongest relative result is SuperCLUE Safety (traditional_safety, #10 of 31); its weakest is Do-Not-Answer (human_harmlessness_rate, #6 of 6).
Freedom. ChatGLM2 6B has an estimated Freedom rank of #38; its 90% source-sensitivity interval is #29–#230. Published Freedom evidence spans 5 eval lineages and 1 of 1 components. Its strongest relative result is Fake Alignment (FINE) (open_ended_safe_response_rate, #1 of 14); its weakest is SuperCLUE Safety (traditional_safety, #19 of 31).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Hugging Face ↗zai-org/chatglm2-6b
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗direct research preview identity date
Safety evals
Rank and direction are specific to the Safety portfolio. Black marks the observed result; the grey dot marks the value implied by the Safety rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| ChiSafetyBenchharmful_response_rate | #12 / 14 | ↓1.08 | Source ↗official | |
| Do-Not-Answerhuman_harmlessness_rate | #6 / 6 | ↑90.95 | Source ↗official | |
| Fake Alignment (FINE)multiple_choice_safe_decision_rate | #13 / 14 | ↑17.33 | Source ↗official | |
| Fake Alignment (FINE)open_ended_safe_response_rate | #13 / 14 | ↑85.33 | Source ↗official | |
| FLAMESdata_protection | #6 / 13 | ↑43.42 | Source ↗official | |
| FLAMESfairness | #10 / 13 | ↑31.73 | Source ↗official | |
| FLAMESlegality | #12 / 13 | ↑28.26 | Source ↗official | |
| FLAMESmorality | #8 / 13 | ↑43.28 | Source ↗official | |
| FLAMESsafety | #11 / 13 | ↑22.61 | Source ↗official | |
| SafetyBenchEM | #10 / 21 | ↑69.4 | Source ↗official | |
| SafetyBenchIA | #10 / 21 | ↑78.2 | Source ↗official | |
| SafetyBenchMH | #10 / 21 | ↑82 | Source ↗official | |
| SafetyBenchOFF | #10 / 21 | ↑68.1 | Source ↗official | |
| SafetyBenchPH | #12 / 21 | ↑67.9 | Source ↗official | |
| SafetyBenchPP | #11 / 21 | ↑76 | Source ↗official | |
| SafetyBenchUB | #12 / 21 | ↑61.6 | Source ↗official | |
| SuperCLUE Safetyinstruction_attack | #20 / 31 | ↑58.62 | Source ↗official | |
| SuperCLUE Safetyresponsible_ai | #24 / 31 | ↑47.27 | Source ↗official | |
| SuperCLUE Safetytraditional_safety | #10 / 31 | ↑78.72 | Source ↗official |
Freedom evals
Rank and direction are specific to the Freedom portfolio. Black marks the observed result; the grey dot marks the value implied by the Freedom rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| ChiSafetyBenchharmful_response_rate | #2 / 14 | ↑1.08 | Source ↗official | |
| ChiSafetyBenchrefusal_rr1 | #5 / 14 | ↓71.43 | Source ↗official | |
| ChiSafetyBenchrefusal_rr2 | #5 / 14 | ↓71.21 | Source ↗official | |
| Do-Not-Answerhuman_harmlessness_rate | #1 / 6 | ↓90.95 | Source ↗official | |
| Fake Alignment (FINE)open_ended_safe_response_rate | #1 / 14 | ↓85.33 | Source ↗official | |
| FLAMESdata_protection | #7 / 13 | ↓43.42 | Source ↗official | |
| FLAMESfairness | #4 / 13 | ↓31.73 | Source ↗official | |
| FLAMESlegality | #1 / 13 | ↓28.26 | Source ↗official | |
| FLAMESmorality | #6 / 13 | ↓43.28 | Source ↗official | |
| FLAMESsafety | #3 / 13 | ↓22.61 | Source ↗official | |
| SuperCLUE Safetyinstruction_attack | #9 / 31 | ↓58.62 | Source ↗official | |
| SuperCLUE Safetytraditional_safety | #19 / 31 | ↓78.72 | Source ↗official |