openai
OpenAI: gpt-oss-safeguard-20b
openai/gpt-oss-safeguard-20b
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
openai/gpt-oss-safeguard-20b is an open-weight, text-to-text safety reasoning model derived from gpt-oss-20b. The Hugging Face documentation describes a 21B-parameter MoE with 3.6B active parameters, Apache-2.0 licensing, a required Harmony response format, and configurable low, medium, or high reasoning effort. It is intended to classify content against policies supplied by the developer and return reviewable reasoning, while OpenRouter lists a 131072-token context and mandatory reasoning. The sources report intended capabilities but provide no independent performance results.[1][2]
Strengths
- Interprets a developer-written safety policy to classify text and support LLM input-output filtering, online labeling, and offline trust-and-safety review.[2]
- Exposes its reasoning for inspection and lets callers choose low, medium, or high reasoning effort to balance review depth and latency; these are vendor-described controls rather than measured results.[2]
- Combines open weights, Apache-2.0 licensing, a 21B MoE design with 3.6B active parameters, documented 16GB VRAM fit, and a 131072-token catalog context.[1][2]
Limitations
- It is intended for safety work; the documentation recommends other gpt-oss models for other applications and says raw chain-of-thought is not for general users.[2]
- The Harmony response format is required; the documentation says the model will not work correctly otherwise, and OpenRouter lists text as its only input and output modality.[1][2]
- The pack contains vendor documentation plus an OpenRouter catalog entry, but no independent evaluations or operational measurements; accuracy, latency, and policy coverage should be validated on the target workload.[1][2]
Where to use it
- Could suit policy-driven pre- or post-generation filtering for LLM traffic when a reviewable rationale is useful.[2]
- Could suit trust-and-safety labeling workflows, near-real-time or offline, where teams apply custom written policies.[2]
- Could suit evaluation or self-hosted deployment of the 20B safety classifier in roughly 16GB GPU-VRAM environments, after confirming Harmony support.[2]
Sources
- [1] OpenRouter catalog · OpenAI: gpt-oss-safeguard-20bSources checked: Nov 13, 2026
- [2] Model documentation · OpenAI: gpt-oss-safeguard-20bSources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
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
- Model ID
- openai/gpt-oss-safeguard-20b
- Creator
- openai
- Added to OpenRouter
- 29.10.2025
- Context length
- 131,072
- Modality
- text->text
- Input modalities
- text
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $0.075 / out $0.300
- Knowledge cutoff
- —
- Expiration date
- —
Tasks and capabilities
A grade appears only where the source supplies a matching measurement for that task, the model is active and the cohort holds at least 10 comparable models. The capability column lists facts the source declares — supported formats, reasoning and tool parameters — which are not measured quality.
| Task | Grade | Source-declared capability |
|---|---|---|
| General quality | No measurement | reasoning: mandatory · parameters: include_reasoning, reasoning |
| Coding | No measurement | — |
| Agents | No measurement | parameters: structured_outputs, tool_choice, tools |
| Writing | No measurement | output: text |
| Reasoning | No measurement | reasoning: mandatory · parameters: include_reasoning, reasoning |
| Document analysis | No measurement | — |
| Vision | No measurement | — |
| Image generation | No measurement | — |
| Audio | No measurement | — |
| Video | No measurement | — |
| Embeddings | No measurement | — |
Evaluations
No evaluations are supplied for this model.
How grades work
Each letter grade compares a model only against other models scored by the same source and metric in the current snapshot. S means the top 10%, A the next 20%, B the next 30%, C the rest. Models without a published measurement are shown as not rated — a supported format alone never implies measured quality.
- S — S — top 10% of the cohort
- A — A — next 20% of the cohort
- B — B — next 30% of the cohort
- C — C — below the B band
Providers
Optional live provider statistics from OpenRouter. Missing values are unknown, not zero.
| Provider | Context length | Price in /1M | Price out /1M | Latency p50 (30m, s) | Throughput p50 (30m, tok/s) | Uptime (30m) |
|---|---|---|---|---|---|---|
| Groq | 131,072 | $0.075 | $0.300 | — | — | 100% |
Provider data retrieved: 13.11.2026
Related articles
No published articles identify this model yet.