Model profile
O3
Evidence summary
O3 has an estimated overall rank of #62; its 90% source-sensitivity interval is #11–#149. Its behavior-only rank is #70; company governance moves the combined estimate to #62. Published evidence spans 23 evals and 6 of 7 behavior components. Its strongest relative result is HELM Safety (harmbench, #3 of 80); its weakest is Social Welfare Function Benchmark (fairness, #19 of 19).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗o3
- OpenRouter ↗openai/o3
- 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 | #221 / 311 | 0.8712 | ↓ lower | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #14 / 80 | 0.845 | ↑ higher | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #56 / 105 | 0.26 | ↑ higher | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #9 / 48 | 47.7 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmask | #19 / 51 | 15.5 | ↓ lower | Source ↗official | |
| CAIS Risk Indexpolitical_manipulation | #6 / 32 | 39 | ↓ lower | Source ↗official | |
| CAIS Risk Indextextquests_harm | #27 / 48 | 18.7 | ↓ lower | Source ↗official | |
| Confabulationsconfabulation_rate | #27 / 52 | 24.75 | ↓ lower | Source ↗official | |
| Emergent Collusionhigh_illegality_game_rate | #9 / 13 | 0.51 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #71 / 260 | 23 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #88 / 260 | 90.33 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #40 / 260 | 93.89 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #166 / 258 | 95.05 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa1_qualified_rate | #5 / 18 | 82.67 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa2_qualified_rate | #1 / 18 | 82.84 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa3_qualified_rate | #5 / 18 | 88.64 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa4_qualified_rate | #5 / 18 | 89.98 | ↑ higher | Source ↗official | |
| FlagEval Safety and Valuesa5_qualified_rate | #8 / 18 | 72.58 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #9 / 49 | 16.01 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #34 / 46 | 7.42 | ↓ lower | Source ↗official | |
| HELM Safetyanthropic_red_team | #50 / 80 | 0.983 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #5 / 80 | 0.979 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #3 / 80 | 0.984 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #32 / 80 | 0.99 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #19 / 80 | 0.973 | ↑ higher | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #34 / 54 | 26.3 | ↑ higher | Source ↗official | |
| MASKlying_probability_pct | #13 / 53 | 16.46 | ↓ lower | Source ↗official | |
| OpenAI o3 and o4-mini System Cardharmful_request_safety | #1 / 2 | 0.92 | ↑ higher | Source ↗official | |
| OpenAI o3 and o4-mini System Cardjailbreak_resistance | #1 / 2 | 0.97 | ↑ higher | Source ↗official | |
| PropensityBenchscore | #1 / 14 | 10.5 | ↓ lower | Source ↗official | |
| Shelleducation_jsr | #7 / 14 | 0.608 | ↓ lower | Source ↗official | |
| Shellfinance_jsr | #5 / 14 | 0.328 | ↓ lower | Source ↗official | |
| Shellmanagement_jsr | #4 / 14 | 0.482 | ↓ lower | Source ↗official | |
| Social Welfare Function Benchmarkfairness | #19 / 19 | 0.433 | ↑ higher | Source ↗official | |
| SOSBenchbiology_pvr | #4 / 23 | 0.156 | ↓ lower | Source ↗official | |
| SOSBenchchemistry_pvr | #3 / 23 | 0.152 | ↓ lower | Source ↗official | |
| SOSBenchmedicine_pvr | #5 / 23 | 0.372 | ↓ lower | Source ↗official | |
| SOSBenchpharmacology_pvr | #5 / 23 | 0.424 | ↓ lower | Source ↗official | |
| SOSBenchphysics_pvr | #3 / 23 | 0.114 | ↓ lower | Source ↗official | |
| SOSBenchpsychology_pvr | #5 / 23 | 0.196 | ↓ 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.