Model profile
Qwen 2.5 14B Instruct
Evidence summary
Qwen 2.5 14B Instruct has an estimated overall rank of #201; its 90% source-sensitivity interval is #103–#227. Its behavior-only rank is #193; company governance moves the combined estimate to #201. Published evidence spans 6 evals and 5 of 7 behavior components. Its strongest relative result is PandaBench JBB direct-request panel (safety_rate, #1 of 46); its weakest is SafeDialBench (legality, #18 of 18).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Hugging Face ↗Qwen/Qwen2.5-14B-Instruct
- 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 |
|---|---|---|---|---|---|
| Agent-SafetyBenchcompromise_availability | #10 / 16 | 29.2 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchharmful_vulnerable_code | #12 / 16 | 29.2 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchleak_sensitive_information | #13 / 16 | 24.4 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchphysical_harm | #11 / 16 | 28 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchproduce_unsafe_information | #12 / 16 | 81.2 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchproperty_loss | #11 / 16 | 31.2 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchspread_unsafe_information | #10 / 16 | 11.2 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchviolate_law_ethics | #12 / 16 | 20.4 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #181 / 260 | 12.92 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #116 / 260 | 89.17 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #135 / 260 | 69.44 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #96 / 258 | 97.73 | ↑ higher | Source ↗official | |
| M3-SafetyBenchoverall_score | #3 / 19 | 95.94 | ↑ higher | Source ↗official | |
| PandaBench JBB direct-request panelsafety_rate | #1 / 46 | 1 | ↑ higher | Source ↗official | |
| SafeDialBenchaggression | #10 / 18 | 7.123 | ↑ higher | Source ↗official | |
| SafeDialBenchethics | #14 / 18 | 7.39 | ↑ higher | Source ↗official | |
| SafeDialBenchfairness | #5 / 18 | 7.56 | ↑ higher | Source ↗official | |
| SafeDialBenchlegality | #18 / 18 | 7.17 | ↑ higher | Source ↗official | |
| SafeDialBenchmorality | #17 / 18 | 7.047 | ↑ higher | Source ↗official | |
| SafeDialBenchprivacy | #15 / 18 | 7.117 | ↑ higher | Source ↗official | |
| SYCON Benchfalse_presupposition_tof | #7 / 11 | 2.31 | ↑ higher | Source ↗official | |
| SYCON Benchunethical_queries_tof | #9 / 11 | 1.14 | ↑ higher | 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.