Model profile
GPT Oss 120B
OpenAIdeveloper
2025-08-05release date
#76 / 267overall rank
25eval lineages
3discovery sources
Evidence summary
GPT Oss 120B has an estimated overall rank of #76; its 90% source-sensitivity interval is #51–#148. Its behavior-only rank is #85; company governance moves the combined estimate to #76. Published evidence spans 25 evals and 7 of 7 behavior components. Its strongest relative result is HELM Safety (harmbench, #1 of 80); its weakest is FORTRESS (over_refusal_score, #45 of 46).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗gpt-oss-120b
- Hugging Face ↗openai/gpt-oss-120b
- NVIDIA ↗qc69jvmznzxy/gpt-oss-120b
- OpenRouter ↗openai/gpt-oss-120b
- Model 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 | #133 / 311 | 0.7815 | ↓ lower | Source ↗official | |
| AgentAbstainabstain | #11 / 17 | 59.5 | ↑ higher | Source ↗official | |
| AgentAbstaincar | #10 / 17 | 58.2 | ↑ higher | Source ↗official | |
| AgentAbstainpaired | #9 / 17 | 46.2 | ↑ higher | Source ↗official | |
| AIRBench 2024 Safety Scenariossafety_scenarios | #5 / 80 | 0.88 | ↑ higher | Source ↗official | |
| Alignment Leaderboardcorrigibility | #16 / 24 | 4.136 | ↑ higher | Source ↗official | |
| Alignment Leaderboardhonesty | #8 / 24 | 3.733 | ↑ higher | Source ↗official | |
| Alignment Leaderboardnon_manipulation | #7 / 24 | 3.611 | ↑ higher | Source ↗official | |
| Alignment Leaderboardrobustness | #3 / 24 | 4.347 | ↑ higher | Source ↗official | |
| Alignment Leaderboardsafety | #16 / 24 | 3.731 | ↑ higher | Source ↗official | |
| Alignment Leaderboardscheming | #10 / 24 | 3.873 | ↑ higher | Source ↗official | |
| ANIMAscore | #12 / 19 | 0.6866 | ↑ higher | Source ↗self run | |
| BrokenMathsycophancy | #2 / 9 | 33.7 | ↓ lower | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #90 / 105 | 0.08 | ↑ higher | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #20 / 105 | 7.245 | ↓ lower | Source ↗official | |
| Confabulationsconfabulation_rate | #24 / 52 | 23.27 | ↓ lower | Source ↗official | |
| DystopiaBenchbasaglia_score | #31 / 50 | 65.67 | ↓ lower | Source ↗official | |
| DystopiaBenchbaudrillard_score | #33 / 50 | 64.03 | ↓ lower | Source ↗official | |
| DystopiaBenchhuxley_score | #33 / 50 | 74.63 | ↓ lower | Source ↗official | |
| DystopiaBenchlaguardia_score | #32 / 50 | 68.63 | ↓ lower | Source ↗official | |
| DystopiaBenchorwell_score | #27 / 50 | 68.8 | ↓ lower | Source ↗official | |
| DystopiaBenchpetrov_score | #35 / 50 | 76.6 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #33 / 260 | 39.79 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #172 / 260 | 86.17 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #74 / 260 | 85.56 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #221 / 258 | 89.09 | ↑ higher | Source ↗official | |
| FORTRESSaverage_risk_score | #1 / 49 | 8.24 | ↓ lower | Source ↗official | |
| FORTRESSover_refusal_score | #45 / 46 | 18 | ↓ lower | Source ↗official | |
| HELM Safetyanthropic_red_team | #18 / 80 | 0.995 | ↑ higher | Source ↗official | |
| HELM Safetybbq | #2 / 80 | 0.985 | ↑ higher | Source ↗official | |
| HELM Safetyharmbench | #1 / 80 | 1 | ↑ higher | Source ↗official | |
| HELM Safetysimple_safety_tests | #1 / 80 | 1 | ↑ higher | Source ↗official | |
| HELM Safetyxstest | #62 / 80 | 0.927 | ↑ higher | Source ↗official | |
| LiveSecBenchethics | #14 / 43 | 61.84 | ↑ higher | Source ↗official | |
| LiveSecBenchfactuality | #22 / 43 | 48.37 | ↑ higher | Source ↗official | |
| LiveSecBenchlegality | #5 / 43 | 86.06 | ↑ higher | Source ↗official | |
| LiveSecBenchprivacy | #13 / 43 | 65.7 | ↑ higher | Source ↗official | |
| LiveSecBenchpsychological_health | #9 / 43 | 71.2 | ↑ higher | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #27 / 31 | 9 | ↓ lower | Source ↗self run | |
| MASKlying_probability_pct | #2 / 53 | 8 | ↓ lower | Source ↗official | |
| ODCV-Benchaverage_severity | #8 / 12 | 1.569 | ↓ lower | Source ↗official | |
| ODCV-Benchmisalignment_rate | #9 / 12 | 36.25 | ↓ lower | Source ↗official | |
| PHAREbias_resistance_diagnostic | #46 / 66 | 0.3884 | ↑ higher | Source ↗official | |
| PHAREhallucination_resistance_diagnostic | #58 / 70 | 0.6608 | ↑ higher | Source ↗official | |
| PHAREharm_resistance_diagnostic | #33 / 70 | 0.9375 | ↑ higher | Source ↗official | |
| PHAREjailbreak_resistance_diagnostic | #16 / 67 | 0.659 | ↑ higher | Source ↗official | |
| SpeciEvalbelief_animal_sentience | #85 / 102 | 6.44 | ↑ higher | Source ↗official | |
| SpeciEvalland_animal_4ns | #46 / 102 | 4.49 | ↓ lower | Source ↗official | |
| SpeciEvalsea_animal_4ns | #77 / 102 | 4.94 | ↓ lower | Source ↗official | |
| SpeciEvalspeciesism | #83 / 102 | 2.57 | ↓ lower | Source ↗official | |
| TACbase_welfare_rate | #25 / 68 | 31.4 | ↑ higher | Source ↗self run | |
| Vigil Mental Health Safetyoverall_score | #15 / 23 | 35 | ↑ 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.