Model profile
Deepseek V3.1
Evidence summary
Deepseek V3.1 has an estimated overall rank of #85; its 90% source-sensitivity interval is #40–#236. Its behavior-only rank is #68; company governance moves the combined estimate to #85. Published evidence spans 12 evals and 7 of 7 behavior components. Its strongest relative result is PHARE (bias_resistance_diagnostic, #8 of 66); its weakest is BrokenMath (sycophancy, #9 of 9).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗deepseek-v3-1
- Hugging Face ↗deepseek-ai/DeepSeek-V3.1
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
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 | #176 / 330 | ↓0.8251 | Source ↗official | |
| BrokenMathsycophancy | #9 / 9 | ↓70.2 | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #96 / 104 | ↓75.17 | Source ↗official | |
| Contextual MoralChoicehuman_agreement | #12 / 22 | ↑0.4 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #78 / 241 | ↑21.19 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #189 / 241 | ↑81.67 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #145 / 241 | ↑58.89 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #171 / 239 | ↑93.45 | Source ↗official | |
| FORTRESSaverage_risk_score | #45 / 49 | ↓60.55 | Source ↗official | |
| FORTRESSover_refusal_score | #10 / 48 | ↓1.76 | Source ↗official | |
| MASKlying_probability_pct | #48 / 53 | ↓53.73 | Source ↗official | |
| PHAREbias_resistance_diagnostic | #8 / 66 | ↑0.6517 | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #65 / 70 | ↑0.6158 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #29 / 70 | ↑0.9443 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #54 / 67 | ↑0.3817 | Source ↗official | |
| Social Welfare Function Benchmarkfairness | #6 / 19 | ↑0.531 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #99 / 113 | ↑6.37 | Source ↗official | |
| SpeciEvalland_animal_4ns | #19 / 113 | ↓4.22 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #68 / 113 | ↓4.83 | Source ↗official | |
| SpeciEvalspeciesism | #34 / 113 | ↓1.75 | Source ↗official | |
| TACbase_welfare_rate | #21 / 76 | ↑32.1 | Source ↗self run | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #16 / 94 | ↑94.5 | 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.