Model profile
Qwen3 32B
Evidence summary
Qwen3 32B has an estimated overall rank of #143; its 90% source-sensitivity interval is #23–#250. Its behavior-only rank is #135; company governance moves the combined estimate to #143. Published evidence spans 6 evals and 7 of 7 behavior components. Its strongest relative result is TAC (base_welfare_rate, #5 of 68); its weakest is Cisco AI Defense Rolling Single-Turn Leaderboard (single_turn_attack_success_rate, #100 of 105).
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
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 |
|---|---|---|---|---|---|
| Alignment Leaderboardcorrigibility | #18 / 24 | 4.087 | ↑ higher | Source ↗official | |
| Alignment Leaderboardhonesty | #19 / 24 | 3.244 | ↑ higher | Source ↗official | |
| Alignment Leaderboardnon_manipulation | #17 / 24 | 3.084 | ↑ higher | Source ↗official | |
| Alignment Leaderboardrobustness | #20 / 24 | 3.107 | ↑ higher | Source ↗official | |
| Alignment Leaderboardsafety | #21 / 24 | 3.327 | ↑ higher | Source ↗official | |
| Alignment Leaderboardscheming | #20 / 24 | 3.385 | ↑ higher | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #100 / 105 | 77.97 | ↓ lower | Source ↗official | |
| PandaBench JBB direct-request panelsafety_rate | #40 / 46 | 0.94 | ↑ higher | Source ↗official | |
| SOSBenchbiology_pvr | #18 / 23 | 0.802 | ↓ lower | Source ↗official | |
| SOSBenchchemistry_pvr | #21 / 23 | 0.784 | ↓ lower | Source ↗official | |
| SOSBenchmedicine_pvr | #17 / 23 | 0.774 | ↓ lower | Source ↗official | |
| SOSBenchpharmacology_pvr | #20 / 23 | 0.946 | ↓ lower | Source ↗official | |
| SOSBenchphysics_pvr | #18 / 23 | 0.74 | ↓ lower | Source ↗official | |
| SOSBenchpsychology_pvr | #17 / 23 | 0.746 | ↓ lower | Source ↗official | |
| TACbase_welfare_rate | #5 / 68 | 48.1 | ↑ higher | Source ↗self run | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #15 / 25 | 8.7 | ↓ 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.