Model profile
Qwen3 32B
Evidence summary
Safety. Qwen3 32B has an estimated Safety rank of #145; its 90% source-sensitivity interval is #27–#290. Its behavior-only rank is #136; company governance moves the combined estimate to #145. Published Safety evidence spans 9 eval lineages and 7 of 7 components. Its strongest relative result is TAC (base_welfare_rate, #5 of 87); its weakest is Cisco AI Defense Rolling Single-Turn Leaderboard (single_turn_attack_success_rate, #99 of 104).
Freedom. Qwen3 32B has an estimated Freedom rank of #46; its 90% source-sensitivity interval is #34–#262. Published Freedom evidence spans 7 eval lineages and 1 of 1 components. Its strongest relative result is Cisco AI Defense Rolling Single-Turn Leaderboard (single_turn_attack_success_rate, #6 of 104); its weakest is VETO Misfired Alignment (misfired_alignment_rate_pct, #15 of 25).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Hugging Face ↗Qwen/Qwen3-32B
- OpenRouter ↗qwen/qwen3-32b
- Model collection ↗Family-level model document · Qwen · first party
- Release source ↗direct research preview identity date
Finetunes
Direct finetunes linked to this canonical base model. Quantized and repackaged derivatives are excluded.
huihui-ai/Qwen3-32B-abliterated ↗Direct non-quantized finetune documented by Hugging Face metadata
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 |
|---|---|---|---|---|
| Adversarial Poetry — AILuminate Baseline and Poetry ASRbaseline_asr | #17 / 24 | ↓9.67 | Source ↗official | |
| Adversarial Poetry — AILuminate Baseline and Poetry ASRpoetry_asr | #20 / 24 | ↓69.05 | Source ↗official | |
| Alignment Leaderboardcorrigibility | #18 / 24 | ↑4.087 | Source ↗official | |
| Alignment Leaderboardhonesty | #19 / 24 | ↑3.244 | Source ↗official | |
| Alignment Leaderboardnon_manipulation | #17 / 24 | ↑3.084 | Source ↗official | |
| Alignment Leaderboardrobustness | #20 / 24 | ↑3.107 | Source ↗official | |
| Alignment Leaderboardsafety | #21 / 24 | ↑3.327 | Source ↗official | |
| Alignment Leaderboardscheming | #20 / 24 | ↑3.385 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #99 / 104 | ↓77.97 | Source ↗official | |
| FinEval 6.0 Safety Awarenesssafety_awareness_score | #3 / 8 | ↑81.93 | Source ↗official | |
| PandaBench JBB direct-request panelsafety_rate | #40 / 46 | ↑0.94 | Source ↗official | |
| SOSBenchbiology_pvr | #18 / 23 | ↓0.802 | Source ↗official | |
| SOSBenchchemistry_pvr | #21 / 23 | ↓0.784 | Source ↗official | |
| SOSBenchmedicine_pvr | #17 / 23 | ↓0.774 | Source ↗official | |
| SOSBenchpharmacology_pvr | #20 / 23 | ↓0.946 | Source ↗official | |
| SOSBenchphysics_pvr | #18 / 23 | ↓0.74 | Source ↗official | |
| SOSBenchpsychology_pvr | #17 / 23 | ↓0.746 | Source ↗official | |
| TACbase_welfare_rate | #5 / 87 | ↑48.1 | Source ↗self run | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #22 / 94 | ↑94.1 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #15 / 25 | ↓8.7 | 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 |
|---|---|---|---|---|
| Adversarial Poetry — AILuminate Baseline and Poetry ASRbaseline_asr | #8 / 24 | ↑9.67 | Source ↗official | |
| Adversarial Poetry — AILuminate Baseline and Poetry ASRpoetry_asr | #5 / 24 | ↑69.05 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #6 / 104 | ↑77.97 | Source ↗official | |
| PandaBench JBB direct-request panelsafety_rate | #6 / 46 | ↓0.94 | Source ↗official | |
| SOSBenchbiology_pvr | #6 / 23 | ↑0.802 | Source ↗official | |
| SOSBenchchemistry_pvr | #3 / 23 | ↑0.784 | Source ↗official | |
| SOSBenchmedicine_pvr | #7 / 23 | ↑0.774 | Source ↗official | |
| SOSBenchpharmacology_pvr | #4 / 23 | ↑0.946 | Source ↗official | |
| SOSBenchphysics_pvr | #6 / 23 | ↑0.74 | Source ↗official | |
| SOSBenchpsychology_pvr | #7 / 23 | ↑0.746 | Source ↗official | |
| SpeechMap model completioncomplete_pct | #12 / 181 | ↑89.8 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_adherence_score | #44 / 156 | ↑4.5 | Source ↗official | |
| UGI Leaderboard — base-model willingnesswillingness_direct_score | #79 / 156 | ↑3 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #15 / 25 | ↓8.7 | 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.