Model profile
GPT 4O Mini
OpenAIdeveloper
2024-07-18release date
#211 / 267overall rank
24eval lineages
Evidence summary
GPT 4O Mini has an estimated overall rank of #211; its 90% source-sensitivity interval is #148–#221. Its behavior-only rank is #219; company governance moves the combined estimate to #211. Published evidence spans 24 evals and 7 of 7 behavior components. Its strongest relative result is Contextual MoralChoice (human_agreement, #1 of 22); its weakest is BullshitBench v2 (clear_pushback_rate, #104 of 105).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gpt-4o-mini
- OpenRouter ↗openai/gpt-4o-mini
- System card ↗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 |
|---|---|---|---|---|---|
| Agent-SafetyBenchcompromise_availability | #13 / 16 | 23.6 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchharmful_vulnerable_code | #13 / 16 | 25.2 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchleak_sensitive_information | #11 / 16 | 28 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchphysical_harm | #12 / 16 | 24.4 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchproduce_unsafe_information | #5 / 16 | 98.4 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchproperty_loss | #12 / 16 | 30 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchspread_unsafe_information | #14 / 16 | 6.8 | ↑ higher | Source ↗official | |
| Agent-SafetyBenchviolate_law_ethics | #14 / 16 | 13.2 | ↑ higher | Source ↗official | |
| AgentDojotargeted_attack_success_rate | #13 / 15 | 0.2719 | ↓ lower | Source ↗official | |
| AgentDojoutility_under_attack | #5 / 15 | 0.4992 | ↑ higher | Source ↗official | |
| AgentHarmharm_score | #10 / 12 | 62.5 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatchild_sexual_exploitation | #14 / 32 | 17.6 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatdefamation | #19 / 32 | 17.4 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chathate | #20 / 32 | 10.9 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatindiscriminate_weapons_cbrne | #13 / 32 | 19.9 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatintellectual_property | #14 / 32 | 10.1 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatnon_violent_crimes | #15 / 32 | 15.7 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatprivacy | #13 / 32 | 10.5 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatsex_related_crimes | #19 / 32 | 16.9 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatsexual_content | #26 / 32 | 42.9 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatspecialized_advice | #19 / 32 | 18.2 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatsuicide_self_harm | #20 / 32 | 15.5 | ↓ lower | Source ↗official | |
| AILuminate General Purpose AI Chatviolent_crimes | #17 / 32 | 18 | ↓ lower | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #57 / 80 | 0.563 | ↑ higher | Source ↗official | |
| AnimalHarmBenchscore | #9 / 10 | 0.002 | ↑ higher | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #104 / 105 | 0.02 | ↑ higher | Source ↗official | |
| CASE-Benchagreement_accuracy | #6 / 7 | 80 | ↑ higher | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #71 / 105 | 54.99 | ↓ lower | Source ↗official | |
| Confabulationsconfabulation_rate | #50 / 52 | 60.89 | ↓ lower | Source ↗official | |
| Contextual MoralChoicehuman_agreement | #1 / 22 | 0.65 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #165 / 260 | 13.7 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #54 / 260 | 92 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #156 / 260 | 60.56 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #80 / 258 | 98 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #34 / 49 | 48.07 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #14 / 46 | 2.27 | ↓ lower | Source ↗official | |
| HELM Safetyanthropic_red_team | #50 / 80 | 0.983 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #59 / 80 | 0.882 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #28 / 80 | 0.849 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #52 / 80 | 0.978 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #39 / 80 | 0.96 | ↑ higher | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #19 / 39 | 1.076 | ↓ lower | Source ↗official | |
| PandaBench JBB direct-request panelsafety_rate | #25 / 46 | 0.98 | ↑ higher | Source ↗official | |
| PHAREbias_resistance_diagnostic | #51 / 66 | 0.379 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #59 / 70 | 0.6597 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #67 / 70 | 0.7729 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #30 / 67 | 0.5214 | ↑ higher | Source ↗official | |
| SafeArenanormalized_safety_score | #2 / 5 | 35.7 | ↑ higher | Source ↗official | |
| SM-Benchadversarial | #55 / 73 | 78.54 | ↑ higher | Source ↗official | |
| SM-Benchambiguous_interpretation | #68 / 73 | 66.07 | ↑ higher | Source ↗official | |
| SM-Benchanti_hallucination | #71 / 73 | 59.16 | ↑ higher | Source ↗official | |
| SM-Bencheq_boundaries | #10 / 73 | 71.35 | ↑ higher | Source ↗official | |
| SM-Benchoverfit | #46 / 73 | 65.03 | ↑ higher | Source ↗official | |
| SORRY-Benchavg | #34 / 51 | 0.36 | ↓ lower | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #97 / 102 | 6.28 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #48 / 102 | 4.53 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #41 / 102 | 4.7 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #85 / 102 | 2.6 | ↓ 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.
ValueCompass
| Dimension | Value | Distribution |
|---|---|---|
| Universalism | 67.2 | |
| Self-direction | 46.2 | |
| Care / Harm | 30.1 | |
| Fairness / Cheating | 28.5 | |
| Ethical | 90.7 |