Model profile
Qwen2.5 72B Instruct
Evidence summary
Qwen2.5 72B Instruct has an estimated overall rank of #94; its 90% source-sensitivity interval is #23–#247. Its behavior-only rank is #82; company governance moves the combined estimate to #94. Published evidence spans 19 evals and 7 of 7 behavior components. Its strongest relative result is HELM Safety (simple_safety_tests, #1 of 80); its weakest is VETO Misfired Alignment (misfired_alignment_rate_pct, #22 of 25).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗qwen2-5-72b-instruct
- Hugging Face ↗Qwen/Qwen2.5-72B-Instruct
- OpenRouter ↗qwen/qwen-2.5-72b-instruct
- Model card ↗Exact model document · Qwen · first party
- Release source ↗repository created at proxy
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 | Source |
|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #203 / 330 | ↓0.8546 | Source ↗official | |
| Agent-SafetyBenchcompromise_availability | #5 / 16 | ↑35.2 | Source ↗official | |
| Agent-SafetyBenchharmful_vulnerable_code | #10 / 16 | ↑29.6 | Source ↗official | |
| Agent-SafetyBenchleak_sensitive_information | #8 / 16 | ↑32.8 | Source ↗official | |
| Agent-SafetyBenchphysical_harm | #9 / 16 | ↑29.6 | Source ↗official | |
| Agent-SafetyBenchproduce_unsafe_information | #6 / 16 | ↑97.2 | Source ↗official | |
| Agent-SafetyBenchproperty_loss | #8 / 16 | ↑38 | Source ↗official | |
| Agent-SafetyBenchspread_unsafe_information | #9 / 16 | ↑12 | Source ↗official | |
| Agent-SafetyBenchviolate_law_ethics | #10 / 16 | ↑24 | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #51 / 80 | ↑0.59 | Source ↗official | |
| Confabulationsconfabulation_rate | #39 / 52 | ↓32.18 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #36 / 241 | ↑38.63 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #94 / 241 | ↑89.67 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #107 / 241 | ↑73.62 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #44 / 239 | ↑98.91 | Source ↗official | |
| FinEval Financial Security Knowledgefinancial_security_accuracy_pct | #1 / 19 | ↑81.8 | Source ↗official | |
| FORTRESSaverage_risk_score | #42 / 49 | ↓56.43 | Source ↗official | |
| FORTRESSover_refusal_score | #5 / 48 | ↓1.19 | Source ↗official | |
| HalluVerse-M3 Hallucination Recognitionhallucination_recognition_accuracy | #10 / 14 | ↑69.81 | Source ↗official | |
| HELM Safetyanthropic_red_team | #13 / 80 | ↑0.996 | Source ↗official | |
| HELM Safetybbq | #23 / 80 | ↑0.954 | Source ↗official | |
| HELM Safetyharmbench | #42 / 80 | ↑0.728 | Source ↗official | |
| HELM Safetysimple_safety_tests | #1 / 80 | ↑1 | Source ↗official | |
| HELM Safetyxstest | #11 / 80 | ↑0.979 | Source ↗official | |
| PandaBench JBB direct-request panelsafety_rate | #14 / 46 | ↑0.99 | Source ↗official | |
| SOSBenchbiology_pvr | #15 / 23 | ↓0.68 | Source ↗official | |
| SOSBenchchemistry_pvr | #15 / 23 | ↓0.56 | Source ↗official | |
| SOSBenchmedicine_pvr | #15 / 23 | ↓0.734 | Source ↗official | |
| SOSBenchpharmacology_pvr | #16 / 23 | ↓0.926 | Source ↗official | |
| SOSBenchphysics_pvr | #15 / 23 | ↓0.678 | Source ↗official | |
| SOSBenchpsychology_pvr | #16 / 23 | ↓0.734 | Source ↗official | |
| SYCON Benchfalse_presupposition_tof | #6 / 11 | ↑2.35 | Source ↗official | |
| SYCON Benchunethical_queries_tof | #7 / 11 | ↑1.32 | Source ↗official | |
| TACbase_welfare_rate | #4 / 76 | ↑49.4 | Source ↗self run | |
| ThaiSafetyBenchsafety_score | #4 / 18 | ↑89.01 | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #22 / 25 | ↓11.4 | 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.
UGI Political Values
Taiwan Sovereignty Benchmark Pro
| Dimension | Value | Distribution |
|---|---|---|
| Pro-Taiwan rubric compatibility | 20 | |
| Warning-phrase rate | 10 | |
| Soft-censorship rate | 5 | |
| API-error rate | 0 |