The Orchestrator › Models › OpenAI
OpenAI: gpt-oss-safeguard-20b
gpt-oss-safeguard-20b is a safety reasoning model from OpenAI built upon gpt-oss-20b. This open-weight, 21B-parameter Mixture-of-Experts (MoE) model offers lower latency for safety tasks like content classification, LLM filtering, and trust...
Specifications
| Provider | OpenAI |
| Model ID | openai/gpt-oss-safeguard-20b |
| Context window | 131K tokens (131,072) |
| Modalities | text → text |
| Tokenizer | GPT |
| Router sources | OpenRouter, TrustedRouter, Vercel AI Gateway |
| Released | 2025-10-29 |
| Tracked since | 2026-11-14 |
| Popularity | Not in the ranked set |
| Reasoning | Mandatory |
Pricing (per million tokens)
| Variant | Source | API ID | Input | Output | Context |
|---|---|---|---|---|---|
| standard | OpenRouter | openai/gpt-oss-safeguard-20b | $0.075 | $0.3 | 131K |
| standard | TrustedRouter | openai/gpt-oss-safeguard-20b | $0.074 | $0.211 | 128K |
| standard | Vercel AI Gateway | openai/gpt-oss-safeguard-20b | $0.07 | $0.2 | 128K |
Recorded price history
| Date | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| 2026-11-14 | $0.07 | $0.2 | 128K |
| 2026-11-14 | $0.074 | $0.211 | 128K |
| 2026-11-14 | $0.075 | $0.3 | 131K |
Capabilities
Agentic tool useStructured outputsReasoning (always on)Controllable reasoningNo content storage (attested)Zero data retentionNo training on inputs (attested)
Trust flags come from TrustedRouter's per-provider attestations.
Supported parameters
include_reasoning, max_tokens, reasoning, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_p