Model profile
GPT Oss Safeguard 20B
OpenAIdeveloper
2025-10-29release date
#97 / 267overall rank
4eval lineages
Evidence summary
GPT Oss Safeguard 20B has an estimated overall rank of #97; its 90% source-sensitivity interval is #27–#182. Its behavior-only rank is #108; company governance moves the combined estimate to #97. Published evidence spans 4 evals and 5 of 7 behavior components. Its strongest relative result is Enkrypt AI Safety Leaderboard (bias_attack_non_success_rate, #63 of 260); its weakest is Enkrypt AI Safety Leaderboard (cbrn_attack_non_success_rate, #206 of 260).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Hugging Face ↗openai/gpt-oss-safeguard-20b
- OpenRouter ↗openai/gpt-oss-safeguard-20b
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- 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 |
|---|---|---|---|---|---|
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #31 / 105 | 17.55 | ↓ lower | Source ↗official | |
| DystopiaBenchbasaglia_score | #36 / 50 | 68.63 | ↓ lower | Source ↗official | |
| DystopiaBenchbaudrillard_score | #34 / 50 | 64.37 | ↓ lower | Source ↗official | |
| DystopiaBenchhuxley_score | #26 / 50 | 72 | ↓ lower | Source ↗official | |
| DystopiaBenchlaguardia_score | #39 / 50 | 69.4 | ↓ lower | Source ↗official | |
| DystopiaBenchorwell_score | #35 / 50 | 73.53 | ↓ lower | Source ↗official | |
| DystopiaBenchpetrov_score | #29 / 50 | 73.6 | ↓ lower | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #63 / 260 | 25.58 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #206 / 260 | 82 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #117 / 260 | 74.44 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #116 / 258 | 96.91 | ↑ higher | Source ↗official | |
| TACbase_welfare_rate | #31 / 68 | 28.21 | ↑ higher | Source ↗self run |