Model profile
O3 Mini
Evidence summary
O3 Mini has an estimated overall rank of #159; its 90% source-sensitivity interval is #84–#200. Its behavior-only rank is #165; company governance moves the combined estimate to #159. Published evidence spans 23 evals and 6 of 7 behavior components. Its strongest relative result is OpenAgentSafety (rule_based_safety_vulnerable, #1 of 7); its weakest is BlueBench AttaQ-100 (attaq_harmlessness_reward_pct, #18 of 18).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗o3-mini
- OpenRouter ↗openai/o3-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 | #139 / 311 | 0.7898 | ↓ lower | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #27 / 80 | 0.749 | ↑ higher | Source ↗official | |
| BlueBench AttaQ-100attaq_harmlessness_reward_pct | #18 / 18 | 80.61 | ↑ higher | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #36 / 48 | 88.3 | ↓ lower | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #44 / 49 | 80 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmachiavelli | #44 / 45 | 105 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmask | #41 / 51 | 51.1 | ↓ lower | Source ↗official | |
| CAIS Risk Indextextquests_harm | #8 / 48 | 13.4 | ↓ lower | Source ↗official | |
| Confabulationsconfabulation_rate | #36 / 52 | 28.96 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #109 / 260 | 18.09 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #60 / 260 | 91.67 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #82 / 260 | 84.44 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #161 / 258 | 95.36 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #25 / 49 | 30.05 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #30 / 46 | 5.65 | ↓ lower | Source ↗official | |
| HELM Safetyanthropic_red_team | #47 / 80 | 0.986 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #36 / 80 | 0.941 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #17 / 80 | 0.952 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #32 / 80 | 0.99 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #55 / 80 | 0.941 | ↑ higher | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #52 / 54 | 22.29 | ↑ higher | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #47 / 50 | 105 | ↓ lower | Source ↗official | |
| MASKlying_probability_pct | #42 / 53 | 50.79 | ↓ lower | Source ↗official | |
| OpenAgentSafetyllm_judge_safety_vulnerable | #7 / 7 | 72.73 | ↓ lower | Source ↗official | |
| OpenAgentSafetyrule_based_safety_vulnerable | #1 / 7 | 32.32 | ↓ lower | Source ↗official | |
| OpenAgentSafetysuccessful_completion | #7 / 7 | 17.29 | ↑ higher | Source ↗official | |
| OpenAI o3-mini System Cardharmful_request_safety | #1 / 2 | 0.9 | ↑ higher | Source ↗official | |
| OpenAI o3-mini System Cardjailbreak_resistance | #1 / 2 | 0.73 | ↑ higher | Source ↗official | |
| PandaBench JBB direct-request panelsafety_rate | #25 / 46 | 0.98 | ↑ higher | Source ↗official | |
| PropensityBenchscore | #5 / 14 | 33.2 | ↓ lower | Source ↗official | |
| SafeDialBenchaggression | #14 / 18 | 7.02 | ↑ higher | Source ↗official | |
| SafeDialBenchethics | #13 / 18 | 7.403 | ↑ higher | Source ↗official | |
| SafeDialBenchfairness | #6 / 18 | 7.557 | ↑ higher | Source ↗official | |
| SafeDialBenchlegality | #17 / 18 | 7.193 | ↑ higher | Source ↗official | |
| SafeDialBenchmorality | #18 / 18 | 7.007 | ↑ higher | Source ↗official | |
| SafeDialBenchprivacy | #17 / 18 | 7.077 | ↑ higher | Source ↗official | |
| SYCON Benchfalse_presupposition_tof | #2 / 11 | 2.98 | ↑ higher | Source ↗official | |
| SYCON Benchunethical_queries_tof | #4 / 11 | 2.31 | ↑ 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.