Model profile
Deepseek V3.2 Exp
Evidence summary
Deepseek V3.2 Exp has an estimated overall rank of #149; its 90% source-sensitivity interval is #18–#236. Its behavior-only rank is #139; company governance moves the combined estimate to #149. Published evidence spans 3 evals and 4 of 7 behavior components. Its strongest relative result is UAVBench safety-critical decision recognition (ethical_safety_critical_accuracy, #1 of 27); its weakest is SpeciEval (belief_animal_sentience, #99 of 102).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Hugging Face ↗deepseek-ai/DeepSeek-V3.2-Exp
- OpenRouter ↗deepseek/deepseek-v3.2-exp
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- 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 | Better | Source |
|---|---|---|---|---|---|
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #71 / 260 | 23 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #209 / 260 | 81.5 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #202 / 260 | 46.67 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #172 / 258 | 94.55 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #99 / 102 | 6.23 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #67 / 102 | 4.7 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #63 / 102 | 4.85 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #40 / 102 | 1.9 | ↓ lower | Source ↗official | |
| UAVBench safety-critical decision recognitionethical_safety_critical_accuracy | #1 / 27 | 0.775 | ↑ 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.