Model profile
Deepseek V4 Pro
Evidence summary
Deepseek V4 Pro has an estimated overall rank of #135; its 90% source-sensitivity interval is #49–#257. Its behavior-only rank is #122; company governance moves the combined estimate to #135. Published evidence spans 16 evals and 7 of 7 behavior components. Its strongest relative result is ANIMA (score, #1 of 22); its weakest is AgentAbstain (abstain, #17 of 17).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗deepseek-v4-pro
- Hugging Face ↗deepseek-ai/DeepSeek-V4-Pro
- NVIDIA ↗qc69jvmznzxy/deepseek-v4-pro
- OpenRouter ↗deepseek/deepseek-v4-pro
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗repository created at proxy
- huggingface.co ↗
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 | #229 / 330 | ↓0.8785 | Source ↗official | |
| AgentAbstainabstain | #17 / 17 | ↑42.8 | Source ↗official | |
| AgentAbstaincar | #16 / 17 | ↑42.3 | Source ↗official | |
| AgentAbstainpaired | #15 / 17 | ↑36.9 | Source ↗official | |
| ANIMAscore | #1 / 22 | ↑0.767 | Source ↗official | |
| Arena Factuality — Text Arena (factuality-only weighting)factuality_bt_rating | #56 / 112 | ↑1437.0 | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #78 / 106 | ↑0.14 | Source ↗official | |
| DystopiaBenchbasaglia_score | #26 / 50 | ↓64.6 | Source ↗official | |
| DystopiaBenchbaudrillard_score | #31 / 50 | ↓62.2 | Source ↗official | |
| DystopiaBenchhuxley_score | #41 / 50 | ↓76.97 | Source ↗official | |
| DystopiaBenchlaguardia_score | #42 / 50 | ↓69.8 | Source ↗official | |
| DystopiaBenchorwell_score | #40 / 50 | ↓74 | Source ↗official | |
| DystopiaBenchpetrov_score | #36 / 50 | ↓77 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #136 / 241 | ↑14.73 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #183 / 241 | ↑82.67 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #92 / 241 | ↑77.78 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #179 / 239 | ↑92.55 | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #20 / 54 | ↑27.61 | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #29 / 33 | ↓9 | Source ↗official | |
| Manager Coercion Benchfabrication_rate | #1 / 15 | ↓0 | Source ↗official | |
| Olam Social Poker — Social Lie Ratesocial_lie_rate_per_10000_turns | #10 / 19 | ↓9 | Source ↗official | |
| Opposite-Narrator Sycophancysycophancy_rate_pct | #17 / 24 | ↓5.1 | Source ↗official | |
| PHAREbias_resistance_diagnostic | #66 / 66 | ↑0.1714 | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #21 / 70 | ↑0.7964 | Source ↗official | |
| PHAREharm_resistance_diagnostic | #22 / 70 | ↑0.9541 | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #44 / 67 | ↑0.4321 | Source ↗official | |
| SM-Benchadversarial | #43 / 79 | ↑80.98 | Source ↗official | |
| SM-Benchambiguous_interpretation | #71 / 79 | ↑68.75 | Source ↗official | |
| SM-Benchanti_hallucination | #51 / 79 | ↑88.48 | Source ↗official | |
| SM-Bencheq_boundaries | #39 / 79 | ↑65.17 | Source ↗official | |
| SM-Benchoverfit | #35 / 79 | ↑76.5 | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #57 / 113 | ↑6.8 | Source ↗official | |
| SpeciEvalland_animal_4ns | #32 / 113 | ↓4.35 | Source ↗official | |
| SpeciEvalsea_animal_4ns | #36 / 113 | ↓4.65 | Source ↗official | |
| SpeciEvalspeciesism | #36 / 113 | ↓1.77 | Source ↗official | |
| TACbase_welfare_rate | #55 / 76 | ↑23.08 | Source ↗official | |
| Vectara HHEM Factual Consistencyfactual_consistency_rate | #39 / 94 | ↑91.4 | 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.