# Model and reasoning controls: https://huggingface.co/openai/gpt-oss-safeguard-20b # Text-only, reasoning and structured outputs: https://openai.com/index/gpt-oss-safeguard-technical-report/ # Tool-call format: https://huggingface.co/openai/gpt-oss-safeguard-20b/blob/main/chat_template.jinja # Context: https://huggingface.co/openai/gpt-oss-safeguard-20b/blob/8a11e17b25c973a24099d4016bf2e17dd7ec1574/config.json # Output is the configured decoder-context ceiling, not a published hosted-API output maximum. # Prompt, reasoning and final output share the 131,072-token budget; usable output is the remaining context. # No separate fixed output cap: https://huggingface.co/openai/gpt-oss-safeguard-20b/blob/8a11e17b25c973a24099d4016bf2e17dd7ec1574/generation_config.json # Reference generation has a caller-supplied token cap (0 = uncapped): https://github.com/openai/gpt-oss/blob/main/gpt_oss/torch/model.py # Release: https://openai.com/index/introducing-gpt-oss-safeguard/ name = "GPT OSS Safeguard 20B" description = "Safety model for policy screening, moderation, and risk-aware routing workflows" family = "gpt-oss" release_date = "2025-10-29" last_updated = "2025-10-29" attachment = false reasoning = true temperature = true tool_call = true structured_output = true open_weights = true [limit] context = 131_072 output = 131_072 [modalities] input = ["text"] output = ["text"] [[weights]] label = "Hugging Face" url = "https://huggingface.co/openai/gpt-oss-safeguard-20b"