Skip to content

gpt-oss models

OpenAI's open-weight reasoning models: gpt-oss-20b runs on a 16 GB GPU and gpt-oss-120b fits on a single 80 GB card. Both are Apache 2.0.

  • Open weights
  • Models 2
  • Context up to 128K
  • Updated Nov 2026
Hugging Face Compare
On this page

At a glance

Type
Open-weight models
Developer
OpenAI (United States)
First release
Aug 2025
Models tracked
2 (2 current)
Longest context
128K tokens
Abilities
Tool callingReasoning mode
Licence
Apache 2.0 plus a short usage policy, so commercial use is allowed.
Smallest local option
gpt-oss-safeguard 20B · ~16 GB
Website
openai.com
Hugging Face
huggingface.co/openai

Newest model

gpt-oss-safeguard

Released Oct 2025 · 128K context

Smallest option

gpt-oss-safeguard 20B

~16 GB VRAM · 14 GB download

gpt-oss models and versions

ModelReleasedSizesContextAbilitiesLicenceStatus
gpt-oss-safeguard Safety classifier that follows your own written policyOct 2025 20B · 120B 20B, 120B128KToolsReasoningApache 2.0 Current
gpt-oss Reasoning and tool use; 20b fits 16 GB, 120b fits one 80 GB GPUAug 2025 20B · 120B 21B (3.6B active), 117B (5.1B active)128KToolsReasoningApache 2.0 Current

Sizes, context windows and abilities come from the Ollama library and Hugging Face model cards, checked 14 Nov 2026. Release months are the first public upload. Spotted a mistake? Tell us.

How much VRAM does gpt-oss need?

Download sizes per Ollama build and the GPU memory each needs with an 8K context. On a CPU-only server the same number is the RAM you need. Longer contexts need more memory; work it out with the LLM VRAM calculator.

Model4-bit (default)8-bit16-bitVRAM (4-bit)Fits onCommand
gpt-oss-safeguard 20B14 GB——~16 GB16 GB GPU
gpt-oss-safeguard 120B65 GB——~72 GB80 GB GPU
gpt-oss 20B14 GB——~16 GB16 GB GPU
gpt-oss 120B65 GB——~72 GB80 GB GPU

VRAM = 4-bit download + about 10% runtime overhead + 0.5 GB for an 8K context, rounded up. Mixture-of-experts (MoE) models still need all their weights in memory, but they run much faster than a dense model of the same size, so they also work well split between GPU and system RAM.

Run gpt-oss-safeguard locally

# Install Ollama (Linux), then pull and chat
curl -fsSL https://ollama.com/install.sh | sh
ollama run gpt-oss-safeguard:20b

New to Ollama? Follow Install Ollama on Ubuntu 26.04 (or 24.04) for GPU drivers, the systemd service and remote access.

What gpt-oss is good for

Agents and tool calling

Trained for function calling, web browsing and Python tool use, with adjustable reasoning effort.

Private reasoning on one GPU

gpt-oss-20b gives strong reasoning in about 15 GB.

Content moderation

gpt-oss-safeguard classifies text against a policy you write in plain language.

Strengths and limits

Strengths

  • Apache 2.0 from OpenAI.
  • Ships natively in 4-bit MXFP4, so the published weights are already small.
  • Reasoning effort can be set to low, medium or high.

Watch out for

  • Text only: no image input.
  • 128K context, shorter than newer families.
  • Has not been updated since 2025; newer models beat it on coding.

About gpt-oss

gpt-oss is OpenAI’s first open-weight language model release since GPT-2. Released in August 2025 under Apache 2.0, it comes in two mixture-of-experts sizes: gpt-oss-20b (21B total, 3.6B active) and gpt-oss-120b (117B total, 5.1B active). The weights ship in MXFP4 4-bit format, which is why the 120B model fits on a single 80 GB GPU.

Both models reason in the same way as OpenAI’s o-series (with low, medium and high effort) and are trained for tool use: function calls, browsing and running Python. In October 2025 OpenAI added gpt-oss-safeguard, a version tuned to classify content against a safety policy you supply at run time.

Guides for running gpt-oss

Tool LLM VRAM Calculator: How Much GPU Memory Do You Need? Updated Nov 2026 AI & LLMs How to Install Ollama on Ubuntu 26.04 and Run LLMs Locally Updated Nov 2026 AI & LLMs How to Install llama.cpp on Ubuntu 26.04 (CPU and NVIDIA GPU) Updated Nov 2026 AI & LLMs How to Install vLLM on Ubuntu 26.04 (OpenAI-Compatible LLM Server) Updated Nov 2026 AI & LLMs How to Install Open WebUI on Ubuntu 26.04 (Self-Hosted ChatGPT Alternative) Updated Nov 2026

Frequently asked questions

What is the latest gpt-oss model?

The newest gpt-oss model is gpt-oss-safeguard, released in Oct 2025. It has a 128K-token context window and supports tool calling, reasoning mode.

How much VRAM do I need to run gpt-oss locally?

The smallest option, gpt-oss-safeguard 20B, is a 14 GB download at 4-bit and needs about 16 GB of GPU memory (or RAM on a CPU-only machine) with an 8K context. Larger sizes need more; see the hardware table on this page, and use the VRAM calculator for longer contexts.

How do I run gpt-oss with Ollama?

Install Ollama, then run "ollama run gpt-oss-safeguard:20b". Ollama downloads the 4-bit build and starts a chat; the same model is also available on the local API at http://localhost:11434.

Can I use gpt-oss commercially?

Apache 2.0 plus a short usage policy, so commercial use is allowed. Always read the licence of the exact model you deploy; it is linked in the models table.

Is gpt-oss the same as ChatGPT?

No. gpt-oss is a separate open-weight model from OpenAI. ChatGPT and the GPT API use proprietary models (see the GPT page) that you cannot download.

Compare gpt-oss

← All AI models

Back to top