Model profile
Deepseek R1
DeepSeekdeveloper
2025-01-20release date
#136 / 267overall rank
39eval lineages
Evidence summary
Deepseek R1 has an estimated overall rank of #136; its 90% source-sensitivity interval is #79–#192. Its behavior-only rank is #123; company governance moves the combined estimate to #136. Published evidence spans 39 evals and 7 of 7 behavior components. Its strongest relative result is FORTRESS (over_refusal_score, #1 of 46); its weakest is Anthropic Agentic Misalignment — lethal action (misaligned_action_rate_pct, #10 of 10).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗deepseek-r1
- Hugging Face ↗deepseek-ai/DeepSeek-R1
- OpenRouter ↗deepseek/deepseek-r1
- 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 |
|---|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #188 / 311 | 0.8405 | ↓ lower | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #62 / 80 | 0.529 | ↑ higher | Source ↗official | |
| Anthropic Agentic Misalignment — blackmailmisaligned_action_rate_pct | #8 / 16 | 79 | ↓ lower | Source ↗official | |
| Anthropic Agentic Misalignment — corporate espionagemisaligned_action_rate_pct | #9 / 16 | 73 | ↓ lower | Source ↗official | |
| Anthropic Agentic Misalignment — lethal actionmisaligned_action_rate_pct | #10 / 10 | 94 | ↓ lower | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #91 / 105 | 0.075 | ↑ higher | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #36 / 43 | 90.9 | ↓ lower | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #45 / 48 | 94.4 | ↓ lower | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #40 / 49 | 73 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmachiavelli | #18 / 45 | 85.1 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmask | #36 / 51 | 42.7 | ↓ lower | Source ↗official | |
| CAIS Risk Indextextquests_harm | #12 / 48 | 15.4 | ↓ lower | Source ↗official | |
| ChiSafetyBenchharmful_response_rate | #8 / 14 | 0.65 | ↓ lower | Source ↗official | |
| ChiSafetyBenchmcq_score | #9 / 12 | 71.41 | ↑ higher | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #99 / 105 | 77.92 | ↓ lower | Source ↗official | |
| Confabulationsconfabulation_rate | #15 / 52 | 15.1 | ↓ lower | Source ↗official | |
| Emergent Collusionhigh_illegality_game_rate | #12 / 13 | 0.71 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #64 / 260 | 25.06 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #258 / 260 | 46.5 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #215 / 260 | 42.22 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #86 / 258 | 97.91 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa1_qualified_rate | #13 / 18 | 77.33 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa2_qualified_rate | #14 / 18 | 76.69 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa3_qualified_rate | #16 / 18 | 82.31 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa4_qualified_rate | #12 / 18 | 84.15 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa5_qualified_rate | #15 / 18 | 69.38 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #49 / 49 | 74.39 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #1 / 46 | 0.49 | ↓ lower | Source ↗official | |
| HELM Safetyanthropic_red_team | #60 / 80 | 0.975 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #17 / 80 | 0.9657 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #71 / 80 | 0.4987 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #51 / 80 | 0.9793 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #38 / 80 | 0.9617 | ↑ higher | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #15 / 54 | 27.99 | ↑ higher | Source ↗official | |
| LiveSecBenchethics | #26 / 43 | 39.41 | ↑ higher | Source ↗official | |
| LiveSecBenchfactuality | #33 / 43 | 32.41 | ↑ higher | Source ↗official | |
| LiveSecBenchlegality | #6 / 43 | 85.68 | ↑ higher | Source ↗official | |
| LiveSecBenchprivacy | #20 / 43 | 59.73 | ↑ higher | Source ↗official | |
| LiveSecBenchpsychological_health | #16 / 43 | 58.87 | ↑ higher | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #18 / 50 | 85.1 | ↓ lower | Source ↗official | |
| MASKlying_probability_pct | #34 / 53 | 44.19 | ↓ lower | Source ↗official | |
| OpenAgentSafetyllm_judge_safety_vulnerable | #5 / 7 | 62.78 | ↓ lower | Source ↗official | |
| OpenAgentSafetyrule_based_safety_vulnerable | #4 / 7 | 32.78 | ↓ lower | Source ↗official | |
| OpenAgentSafetysuccessful_completion | #6 / 7 | 20.86 | ↑ higher | Source ↗official | |
| PandaBench JBB direct-request panelsafety_rate | #42 / 46 | 0.93 | ↑ higher | Source ↗official | |
| PHAREbias_resistance_diagnostic | #64 / 66 | 0.2549 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #39 / 70 | 0.7289 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #27 / 70 | 0.9515 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #50 / 67 | 0.4067 | ↑ higher | Source ↗official | |
| RefusalBenchyouden_j | #13 / 19 | 0.1424 | ↑ higher | Source ↗official | |
| SABERoverall_safety_rate | #13 / 13 | 15.29 | ↑ higher | Source ↗official | |
| SABERscenario_a_safety_rate | #13 / 13 | 15.69 | ↑ higher | Source ↗official | |
| SABERscenario_b_safety_rate | #13 / 13 | 24.12 | ↑ higher | Source ↗official | |
| SABERscenario_c_safety_rate | #13 / 13 | 8.07 | ↑ higher | Source ↗official | |
| SafeDialBenchaggression | #2 / 18 | 7.273 | ↑ higher | Source ↗official | |
| SafeDialBenchethics | #18 / 18 | 7.303 | ↑ higher | Source ↗official | |
| SafeDialBenchfairness | #3 / 18 | 7.607 | ↑ higher | Source ↗official | |
| SafeDialBenchlegality | #12 / 18 | 7.38 | ↑ higher | Source ↗official | |
| SafeDialBenchmorality | #12 / 18 | 7.183 | ↑ higher | Source ↗official | |
| SafeDialBenchprivacy | #13 / 18 | 7.193 | ↑ higher | Source ↗official | |
| Shelleducation_jsr | #9 / 14 | 0.672 | ↓ lower | Source ↗official | |
| Shellfinance_jsr | #8 / 14 | 0.522 | ↓ lower | Source ↗official | |
| Shellmanagement_jsr | #9 / 14 | 0.682 | ↓ lower | Source ↗official | |
| SM-Benchadversarial | #21 / 73 | 84.88 | ↑ higher | Source ↗official | |
| SM-Benchambiguous_interpretation | #66 / 73 | 67.26 | ↑ higher | Source ↗official | |
| SM-Benchanti_hallucination | #58 / 73 | 84.82 | ↑ higher | Source ↗official | |
| SM-Bencheq_boundaries | #54 / 73 | 57.58 | ↑ higher | Source ↗official | |
| SM-Benchoverfit | #34 / 73 | 74.86 | ↑ higher | Source ↗official | |
| Social Welfare Function Benchmarkfairness | #8 / 19 | 0.523 | ↑ higher | Source ↗official | |
| SOSBenchbiology_pvr | #19 / 23 | 0.814 | ↓ lower | Source ↗official | |
| SOSBenchchemistry_pvr | #22 / 23 | 0.834 | ↓ lower | Source ↗official | |
| SOSBenchmedicine_pvr | #18 / 23 | 0.806 | ↓ lower | Source ↗official | |
| SOSBenchpharmacology_pvr | #22 / 23 | 0.964 | ↓ lower | Source ↗official | |
| SOSBenchphysics_pvr | #22 / 23 | 0.872 | ↓ lower | Source ↗official | |
| SOSBenchpsychology_pvr | #19 / 23 | 0.806 | ↓ lower | Source ↗official | |
| SpeciesismBenchexplicit_speciesism_scale | #1 / 7 | 1.84 | ↓ lower | Source ↗official | |
| SpeciesismBenchmorally_wrong_rate | #3 / 8 | 39.88 | ↑ higher | Source ↗official | |
| SpeciesismBenchspeciesism_recognition_rate | #7 / 8 | 76.42 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #68 / 102 | 6.62 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #42 / 102 | 4.47 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #29 / 102 | 4.65 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #61 / 102 | 2.15 | ↓ lower | Source ↗official | |
| SYCON Benchfalse_presupposition_tof | #1 / 11 | 3.21 | ↑ higher | Source ↗official | |
| SYCON Benchunethical_queries_tof | #2 / 11 | 2.72 | ↑ higher | Source ↗official | |
| TrustLLM contemporary collapsed applicationtrustllm | #3 / 8 | 0.62 | ↑ higher | Source ↗official | |
| VETO Misfired Alignmentmisfired_alignment_rate_pct | #1 / 25 | 4.7 | ↓ lower | 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
ValueCompass
| Dimension | Value | Distribution |
|---|---|---|
| Universalism | 69.2 | |
| Self-direction | 51 | |
| Care / Harm | 48.9 | |
| Fairness / Cheating | 47.5 | |
| Ethical | 94.9 |