gpt-oss-safeguard-20b
gpt-oss-safeguard-20b is a downloadable safety-reasoning model derived from gpt-oss. It classifies text against a written policy supplied by the operator.
The supplied policy controls classification
The 20b release can run in a local serving stack. The guide provides deployment examples for the separate 20B and 120B models. Policy definitions and thresholds come from the supplied instructions, rather than one fixed moderation taxonomy.OpenAI and ROOST guide to gpt-oss-safeguard
Review the rule and the decision together
Practical guidance. For a narrow review queue, begin with a concise policy and clear examples of permitted and disallowed content. Compare ambiguous cases separately from obvious ones. Measure the serving resources and review quality before choosing the smaller model for a larger workload.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | gpt-oss-safeguard-20b |
Related models
- gpt-oss-safeguard-120b
OpenAI ยท gpt-oss-safeguard
A downloadable 120B-class safety-reasoning model that classifies text against a supplied policy.
Launch date not confirmedDownloadable