Model profile
O4 Mini
Evidence summary
O4 Mini has an estimated overall rank of #142; its 90% source-sensitivity interval is #75–#176. Its behavior-only rank is #152; company governance moves the combined estimate to #142. Published evidence spans 24 evals and 6 of 7 behavior components. Its strongest relative result is HELM Safety (simple_safety_tests, #1 of 80); its weakest is OpenAI o3 and o4-mini System Card (harmful_request_safety, #2 of 2).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗o4-mini
- OpenRouter ↗openai/o4-mini
- System card ↗Exact 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 | #149 / 311 | 0.8012 | ↓ lower | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #23 / 80 | 0.785 | ↑ higher | Source ↗official | |
| BlueBench AttaQ-100attaq_harmlessness_reward_pct | #12 / 18 | 83.55 | ↑ higher | Source ↗official | |
| BrokenMathsycophancy | #6 / 9 | 46.6 | ↓ lower | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #95 / 105 | 0.06 | ↑ higher | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #24 / 48 | 67.6 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmask | #26 / 51 | 21.4 | ↓ lower | Source ↗official | |
| CAIS Risk Indextextquests_harm | #26 / 48 | 18.6 | ↓ lower | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #28 / 105 | 12.31 | ↓ lower | Source ↗official | |
| Confabulationsconfabulation_rate | #32 / 52 | 26.73 | ↓ lower | Source ↗official | |
| Emergent Collusionhigh_illegality_game_rate | #11 / 13 | 0.62 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #90 / 260 | 20.16 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #148 / 260 | 87.33 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #80 / 260 | 85 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #191 / 258 | 93.27 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa1_qualified_rate | #9 / 18 | 81.28 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa2_qualified_rate | #8 / 18 | 80.18 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa3_qualified_rate | #7 / 18 | 87.75 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa4_qualified_rate | #9 / 18 | 88.24 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa5_qualified_rate | #9 / 18 | 72.2 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #21 / 49 | 21.49 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #26 / 46 | 5.14 | ↓ lower | Source ↗official | |
| HELM Safetyanthropic_red_team | #54 / 80 | 0.982 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #38 / 80 | 0.94 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #11 / 80 | 0.97 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #1 / 80 | 1 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #17 / 80 | 0.974 | ↑ higher | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #43 / 54 | 24.53 | ↑ higher | Source ↗official | |
| MASKlying_probability_pct | #19 / 53 | 24.23 | ↓ lower | Source ↗official | |
| OpenAI o3 and o4-mini System Cardharmful_request_safety | #2 / 2 | 0.9 | ↑ higher | Source ↗official | |
| OpenAI o3 and o4-mini System Cardjailbreak_resistance | #2 / 2 | 0.96 | ↑ higher | Source ↗official | |
| PropensityBenchscore | #3 / 14 | 15.8 | ↓ lower | Source ↗official | |
| Social Welfare Function Benchmarkfairness | #16 / 19 | 0.445 | ↑ higher | Source ↗official | |
| SOSBenchbiology_pvr | #6 / 23 | 0.262 | ↓ lower | Source ↗official | |
| SOSBenchchemistry_pvr | #4 / 23 | 0.206 | ↓ lower | Source ↗official | |
| SOSBenchmedicine_pvr | #7 / 23 | 0.462 | ↓ lower | Source ↗official | |
| SOSBenchpharmacology_pvr | #3 / 23 | 0.408 | ↓ lower | Source ↗official | |
| SOSBenchphysics_pvr | #7 / 23 | 0.22 | ↓ lower | Source ↗official | |
| SOSBenchpsychology_pvr | #8 / 23 | 0.314 | ↓ 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.