gpt-oss-20b vs gpt-oss-120b: Which Open Model Should You Run?

OpenAI shipped two open-weight models. One runs on a laptop, the other needs a serious GPU. Here is exactly how they differ and which one fits your hardware and use case.

6 minBeginner

When OpenAI released gpt-oss, it shipped two sizes: gpt-oss-20b and gpt-oss-120b. Both are Apache 2.0, both are mixture-of-experts, and both run locally — but the hardware they demand and the quality they deliver are very different. This guide breaks down the practical differences so you pick the right one the first time instead of downloading 65 GB you cannot run.

The specs side by side

	gpt-oss-20b	gpt-oss-120b
Total parameters	20.9B	117B
Active per token	3.6B	5.1B
Memory needed	~16 GB	~60–80 GB
Download size	~13 GB	~65 GB
Context window	128K	128K
License	Apache 2.0	Apache 2.0
Runs on	16 GB laptop / GPU / Mac	80 GB GPU or 64 GB+ unified Mac
Ollama tag	gpt-oss:20b	gpt-oss:120b

Why active parameters matter

Both models are MoE, so only a fraction of their weights run on each token (3.6B for the 20b, 5.1B for the 120b). That keeps inference fast on both. But you still have to fit ALL the weights in memory to load the model — which is why the 120b needs 60–80 GB even though it only activates 5.1B at a time.

Choose gpt-oss-20b if…

You have a 16 GB laptop, a 16 GB GPU, or an Apple Silicon Mac with 16–24 GB of unified memory.
You want fast, responsive everyday use: chat, coding help, summarization, drafting.
You are running it alongside other apps and cannot dedicate the whole machine to one model.
You are just getting started with local AI and want the lowest-friction path.

Choose gpt-oss-120b if…

You have an 80 GB datacenter GPU (H100/A100), a multi-GPU rig, or a Mac with 64 GB+ unified memory (M-series Pro/Max/Ultra).
You need the strongest reasoning and coding quality an open OpenAI model can give, and you are willing to trade speed and hardware cost for it.
You are building a shared inference server for a team rather than a personal assistant.

The honest recommendation for most people

Start with gpt-oss-20b. It runs on hardware you probably already own, and for everyday tasks the quality gap is smaller than the size difference suggests. Only move up to the 120b once you have a specific task where the 20b falls short AND the hardware to load it. Most people never need to.

How to run whichever you pick

Both install the same way through Ollama — the only thing that changes is the tag. For the full step-by-step (install, first chat, reasoning control, and the OpenAI-compatible API), follow our guide on running gpt-oss-20b locally with Ollama; swap in gpt-oss:120b if your hardware can handle it.

zsh — pick your size

Runs on a 16 GB machine:

$ollama run gpt-oss:20b

Needs ~60-80 GB of memory:

$ollama run gpt-oss:120b

Local models run better with more VRAM. CompareRTX GPUs on Amazonbefore you upgrade.(affiliate link. We may earn a commission at no extra cost. Disclosure)