Model profile
GPT 5 Mini
OpenAIdeveloper
2025-08-07release date
#20 / 267overall rank
24eval lineages
Evidence summary
GPT 5 Mini has an estimated overall rank of #20; its 90% source-sensitivity interval is #13–#96. Its behavior-only rank is #22; company governance moves the combined estimate to #20. Published evidence spans 24 evals and 7 of 7 behavior components. Its strongest relative result is HELM Safety (simple_safety_tests, #1 of 80); its weakest is FORTRESS (over_refusal_score, #44 of 46).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gpt-5-mini
- OpenRouter ↗openai/gpt-5-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 |
|---|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #37 / 311 | 0.424 | ↓ lower | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #11 / 80 | 0.857 | ↑ higher | Source ↗official | |
| Alignment Leaderboardcorrigibility | #3 / 24 | 4.495 | ↑ higher | Source ↗official | |
| Alignment Leaderboardhonesty | #4 / 24 | 4.157 | ↑ higher | Source ↗official | |
| Alignment Leaderboardnon_manipulation | #5 / 24 | 4.243 | ↑ higher | Source ↗official | |
| Alignment Leaderboardrobustness | #1 / 24 | 4.6 | ↑ higher | Source ↗official | |
| Alignment Leaderboardsafety | #5 / 24 | 4.288 | ↑ higher | Source ↗official | |
| Alignment Leaderboardscheming | #5 / 24 | 4.193 | ↑ higher | Source ↗official | |
| ANIMAscore | #7 / 19 | 0.7177 | ↑ higher | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #27 / 43 | 84.3 | ↓ lower | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #30 / 48 | 78.2 | ↓ lower | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #33 / 49 | 65 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmachiavelli | #13 / 45 | 84.2 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmask | #27 / 51 | 21.7 | ↓ lower | Source ↗official | |
| CAIS Risk Indextextquests_harm | #23 / 48 | 17.5 | ↓ lower | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #18 / 105 | 5.085 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #31 / 260 | 41.09 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #11 / 260 | 96.83 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #54 / 260 | 91.11 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #163 / 258 | 95.2 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #12 / 49 | 17 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #44 / 46 | 14.17 | ↓ lower | Source ↗official | |
| HELM Safetyanthropic_red_team | #32 / 80 | 0.991 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #19 / 80 | 0.963 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #10 / 80 | 0.971 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #1 / 80 | 1 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #14 / 80 | 0.977 | ↑ higher | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #33 / 54 | 26.31 | ↑ higher | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #13 / 50 | 84.2 | ↓ lower | Source ↗official | |
| MASKlying_probability_pct | #14 / 53 | 17.4 | ↓ lower | Source ↗official | |
| MORUscore | #3 / 13 | 81.57 | ↑ higher | Source ↗official | |
| PHAREbias_resistance_diagnostic | #31 / 66 | 0.4641 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #22 / 70 | 0.7958 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #7 / 70 | 0.9829 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #15 / 67 | 0.686 | ↑ higher | Source ↗official | |
| SM-Benchadversarial | #33 / 73 | 82.44 | ↑ higher | Source ↗official | |
| SM-Benchambiguous_interpretation | #8 / 73 | 91.67 | ↑ higher | Source ↗official | |
| SM-Benchanti_hallucination | #52 / 73 | 86.91 | ↑ higher | Source ↗official | |
| SM-Bencheq_boundaries | #63 / 73 | 52.81 | ↑ higher | Source ↗official | |
| SM-Benchoverfit | #69 / 73 | 25.14 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #86 / 102 | 6.43 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #11 / 102 | 4.17 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #25 / 102 | 4.58 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #87 / 102 | 2.65 | ↓ lower | Source ↗official | |
| TACbase_welfare_rate | #37 / 68 | 26.28 | ↑ higher | Source ↗self run |