Decision models · OpenAI

gpt-oss-safeguard 20B

Name: gpt-oss-safeguard 20B
Author: OpenAI

Task: Safety and policy checks
Returns: Reasoned verdict
Precision tier: 76 / 100
Size: 22B (3.6B active)
Max input: 128K tokens
License: apache-2.0
Released: 2025-10

What it decides

Returns: Reasons step by step over a policy you write, then gives a label; reasoning effort is adjustable (low, medium, high).
Checks: Any content against your own policy
Languages: English
Precision: 76 / 100 · Policy-following classifier from OpenAI and ROOST; scored low (F1 0.38) on a fixed-taxonomy benchmark that does not supply a policy (arXiv 2605.28830).
Clarity: Reasoned verdict (10 of 15 clarity points) · writes about 160 tokens before the answer, so it is slower but explains itself.
How it runs: Writes a short answer · a small language model, so it runs in the same apps as LLMs.
License: apache-2.0 (Open (Apache, MIT))

Source: model card on Hugging Face (opens in a new tab) · How decision models are rated

How to run it

Quickest · Ollama

One command downloads a ready-to-run version and starts a chat.

Get Ollama at ollama.com/download (opens in a new tab) · model page (opens in a new tab)

ollama run gpt-oss-safeguard:20b

Quantized files · GGUF

Works in LM Studio, llama.cpp, Jan and Ollama. No sign-in needed.

huggingface.co/unsloth/gpt-oss-safeguard-20b-GGUF (opens in a new tab)

ollama run hf.co/unsloth/gpt-oss-safeguard-20b-GGUF

Original weights · Hugging Face

Full-precision files from OpenAI, for fine-tuning, vLLM, Transformers.

huggingface.co/openai/gpt-oss-safeguard-20b (opens in a new tab)

No sign-in needed.

Links checked 2 days ago against Hugging Face, Ollama and GitHub.

Apps that run gpt-oss-safeguard 20B

Our pick for each kind of computer, assuming the model fits in its memory. Pick your own GPU or device in the Model Finder for an exact answer.

Mac (Apple silicon)

LM Studio

Runs the MLX build, typically 35–90% faster than GGUF for a model with 3.6B active parameters on Apple silicon.

Also good:

Windows PC, NVIDIA GPU

Ollama

One command with Ollama’s own build of this model.

Also good:

Windows PC, AMD Radeon

Lemonade

AMD’s own app: llama.cpp tuned for Radeon on Vulkan and ROCm.

Also good:

Linux, NVIDIA GPU

Ollama

One command with Ollama’s own build of this model.

Also good:

Linux server, 80 GB+ GPUs

vLLM

The full model fits in GPU memory at 8-bit, so it can batch many requests: best for serving a team or an app.

Also good:

App	Type	Opens it as	Systems	Engine	Start
LM Studio	Desktop app	GGUF or MLX	macOSWindowsLinux	llama.cpp, MLX	Open in LM Studio
Ollama	CLI + API	Ollama library	macOSWindowsLinux	llama.cpp, MLX	`ollama run gpt-oss-safeguard:20b`
llama.cpp	CLI + API	GGUF	macOSWindowsLinux	llama.cpp	`llama-server -hf unsloth/gpt-oss-safeguard-20b-GGUF:Q4_K_M`
Jan	Desktop app	GGUF or MLX	macOSWindowsLinux	llama.cpp, MLX	Open in Jan
MLX LM	CLI + API	MLX	macOS	MLX	`mlx_lm.chat --model lmstudio-community/gpt-oss-safeguard-20b-MLX-MXFP4`
vLLM	Server	Original weights	WSLWindowsLinux	vLLM	`vllm serve openai/gpt-oss-safeguard-20b`
SGLang	Server	Original weights	WSLWindowsLinux	SGLang	`python3 -m sglang.launch_server --model-path openai/gpt-oss-safeguard-20b`
Docker Model Runner	CLI + API	GGUF	macOSWindowsLinux	llama.cpp, vLLM	`docker model run hf.co/unsloth/gpt-oss-safeguard-20b-GGUF:Q4_K_M`
Lemonade	Desktop app	GGUF	macOSWindowsLinux	llama.cpp, ONNX Runtime GenAI	`lemonade pull unsloth/gpt-oss-safeguard-20b-GGUF:Q4_K_M`
Unsloth Desktop	Desktop app	GGUF or MLX	macOSWindowsLinux	llama.cpp, MLX, stable-diffusion.cpp	Open in Unsloth Desktop
Atomic Chat	Desktop app	GGUF or MLX	macOSWindowsLinux	llama.cpp, MLX-VLM	Open in Atomic Chat
LocalAI	Server	GGUF	macOSWSLWindowsLinux	llama.cpp, rerankers, Transformers	Get LocalAI
Sentence Transformers	Library	Original weights	macOSWindowsLinux	PyTorch	Get Sentence Transformers

Which apps can open a model depends on the files published for it: GGUF for llama.cpp-based apps, MLX for Apple’s engine, the original weights for vLLM and SGLang. Compare every app.

Memory and speed

Each quantization on an NVIDIA GeForce RTX 4090, for one decision on a 512-token input. The best hardware table below covers other systems and devices.

Format	Memory needed	Time per decision	Decisions per second	Quality	Download
`MXFP4`	12.9 GB	1.0 s	1.0/s	76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking.	Original weights (opens in a new tab)
`F16`	43.7 GB	—	—	76Great quality (A) · 76 of 100: how capable the model is at F16, whatever it runs on. Match score 10/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab) Original weights (opens in a new tab)

One request at a time. Servers that batch (Text Embeddings Inference, vLLM, Infinity) handle many more decisions per second.

Best hardware for gpt-oss-safeguard 20B

The least expensive GPU in our catalog that runs it without offloading is the AMD Radeon RX 7600 XT (launch price $329) at MXFP4.

Quality	Hardware	Quant	Memory	Speed	Runs
76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA RTX PRO 6000 Blackwell	`MXFP4`	17.7 GB / 96.0 GB	808 ms	Slow
76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA GeForce RTX 5090	`MXFP4`	14.5 GB / 32.0 GB	812 ms	Slow
76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA GeForce RTX 4090	`MXFP4`	14.1 GB / 24.0 GB	1.0 s	Slow
76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA GeForce RTX 3090	`MXFP4`	14.1 GB / 24.0 GB	1.1 s	Slow
76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking.	AMD Radeon RX 7900 XTX	`MXFP4`	14.1 GB / 24.0 GB	1.1 s	Slow
76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking.	Apple M3 Ultra (256 GB)	`MXFP4`	28.3 GB / 192 GB	1.3 s	Slow
76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking.	Apple M3 Ultra (512 GB)	`MXFP4`	43.6 GB / 384 GB	1.3 s	Slow
76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking.	Apple M2 Ultra (192 GB)	`MXFP4`	24.4 GB / 144 GB	1.3 s	Slow
76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA GeForce RTX 5080	`MXFP4`	13.7 GB / 16.0 GB	1.0 s	Slowtight fit
76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA GeForce RTX 5070 Ti	`MXFP4`	13.7 GB / 16.0 GB	1.1 s	Slowtight fit

Check it on your own hardware

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.