Model profile
GPT 4.1
OpenAIdeveloper
2025-04-14release date
#55 / 267overall rank
25eval lineages
Evidence summary
GPT 4.1 has an estimated overall rank of #55; its 90% source-sensitivity interval is #20–#136. Its behavior-only rank is #66; company governance moves the combined estimate to #55. Published evidence spans 25 evals and 7 of 7 behavior components. Its strongest relative result is HELM Safety (simple_safety_tests, #1 of 80); its weakest is Vigil Mental Health Safety (overall_score, #22 of 23).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gpt-4-1
- OpenRouter ↗openai/gpt-4.1
- Official model documentation ↗Family-level model document · openai · first party
- 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 | #146 / 311 | 0.7962 | ↓ lower | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #44 / 80 | 0.648 | ↑ higher | Source ↗official | |
| Anthropic Agentic Misalignment — blackmailmisaligned_action_rate_pct | #10 / 16 | 80 | ↓ lower | Source ↗official | |
| Anthropic Agentic Misalignment — corporate espionagemisaligned_action_rate_pct | #12 / 16 | 99 | ↓ lower | Source ↗official | |
| Anthropic Agentic Misalignment — lethal actionmisaligned_action_rate_pct | #4 / 10 | 54 | ↓ lower | Source ↗official | |
| BlueBench AttaQ-100attaq_harmlessness_reward_pct | #6 / 18 | 87.49 | ↑ higher | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #77 / 105 | 0.14 | ↑ higher | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #68 / 105 | 53.49 | ↓ lower | Source ↗official | |
| Contextual MoralChoicehuman_agreement | #3 / 22 | 0.5 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #113 / 260 | 17.83 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #98 / 260 | 90 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #113 / 260 | 75 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #172 / 258 | 94.55 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa1_qualified_rate | #4 / 18 | 84.2 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa2_qualified_rate | #2 / 18 | 82.54 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa3_qualified_rate | #1 / 18 | 90.46 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa4_qualified_rate | #1 / 18 | 92.11 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa5_qualified_rate | #11 / 18 | 71.36 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #37 / 49 | 53.02 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #17 / 46 | 2.87 | ↓ lower | Source ↗official | |
| HELM Safetyanthropic_red_team | #27 / 80 | 0.993 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #45 / 80 | 0.926 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #19 / 80 | 0.917 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #1 / 80 | 1 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #11 / 80 | 0.979 | ↑ higher | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #29 / 54 | 26.83 | ↑ higher | Source ↗official | |
| MASKlying_probability_pct | #38 / 53 | 48.87 | ↓ lower | Source ↗official | |
| PHAREbias_resistance_diagnostic | #20 / 66 | 0.5241 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #28 / 70 | 0.7764 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #41 / 70 | 0.923 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #40 / 67 | 0.461 | ↑ higher | Source ↗official | |
| SM-Benchadversarial | #11 / 73 | 87.32 | ↑ higher | Source ↗official | |
| SM-Benchambiguous_interpretation | #23 / 73 | 87.8 | ↑ higher | Source ↗official | |
| SM-Benchanti_hallucination | #39 / 73 | 91.1 | ↑ higher | Source ↗official | |
| SM-Bencheq_boundaries | #9 / 73 | 73.31 | ↑ higher | Source ↗official | |
| SM-Benchoverfit | #31 / 73 | 76.5 | ↑ higher | Source ↗official | |
| Social Welfare Function Benchmarkfairness | #11 / 19 | 0.483 | ↑ higher | Source ↗official | |
| SOSBenchbiology_pvr | #11 / 23 | 0.374 | ↓ lower | Source ↗official | |
| SOSBenchchemistry_pvr | #9 / 23 | 0.314 | ↓ lower | Source ↗official | |
| SOSBenchmedicine_pvr | #12 / 23 | 0.57 | ↓ lower | Source ↗official | |
| SOSBenchpharmacology_pvr | #13 / 23 | 0.85 | ↓ lower | Source ↗official | |
| SOSBenchphysics_pvr | #9 / 23 | 0.41 | ↓ lower | Source ↗official | |
| SOSBenchpsychology_pvr | #12 / 23 | 0.498 | ↓ lower | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #53 / 102 | 6.78 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #65 / 102 | 4.67 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #59 / 102 | 4.83 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #6 / 102 | 1.31 | ↓ lower | Source ↗official | |
| TACbase_welfare_rate | #11 / 68 | 37.18 | ↑ higher | Source ↗official | |
| UAVBench safety-critical decision recognitionethical_safety_critical_accuracy | #13 / 27 | 0.7 | ↑ higher | Source ↗official | |
| Vigil Mental Health Safetyoverall_score | #22 / 23 | 26 | ↑ 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.