gpt-oss-safeguard technical report
gpt-oss-safeguard technical report
AI summary
gpt-oss-safeguard-120b and gpt-oss-safeguard-20b are two open-weight reasoning models post-trained by OpenAI on gpt-oss that can label content according to a given policy. The technical report describes the capabilities of these two models and provides safety evaluation results using the original gpt-oss model as a baseline.
Why it was selected
The report uses the original gpt-oss model as a baseline and provides safety evaluation results for the two gpt-oss-safeguard variants, making it easy to compare the gap between models that label content according to such policies and the baseline.