How long does it take to run gpt-oss:20b?

gpt-oss is OpenAI's open-weight models designed for powerful reasoning, agentic tasks, and versatile developer use cases. You can easily use it via Ollama and read more about it in their blog post.

Let's see how long it takes to generate some output using the same prompt on some of my computers at home. For this post, I'm using the same version of Ollama and gpt-oss:20b model.

ollama --version

ollama version is 0.11.4

ollama list

NAME               ID              SIZE      MODIFIED     
gpt-oss:20b        e95023cf3b7b    13 GB     2 hours ago

My first computer runs Debian 12 and has a Intel(R) Core(TM) i5-8500 CPU @ 3.00GHz CPU but with no GPU.

date; ollama run gpt-oss:20b "Generate a document on the history of bioinformatics as a field.">output; date

Sun Aug 10 11:45:41 AM JST 2025
Sun Aug 10 12:07:11 PM JST 2025

About 22 minutes.

My second computer runs Ubuntu 24.04.2 LTS, has a Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz CPU and a NVIDIA® GeForce® RTX 2060 SUPER.

date; ollama run gpt-oss:20b "Generate a document on the history of bioinformatics as a field.">output; date

Sun 10 Aug 2025 11:48:16 JST
Sun 10 Aug 2025 11:57:50 JST

About 9 minutes.

My third computer (a laptop) runs Windows 11, has a 12th Gen Intel(R) Core(TM) i7-12650H CPU and a laptop NVIDIA® GeForce® RTX 4060.

I tried using Windows PowerShell to run Ollama but I tried a few times and this was too slow.

Get-Date; ollama run gpt-oss:20b "Generate a document on the history of bioinformatics as a field.">output; Get-Date

Using the Ollama GUI, I entered the some prompt and just manually timed (using Get-Date) how long it took to generate an output.

Sunday, 10 August 2025 1:20:10 PM
Sunday, 10 August 2025 1:30:20 PM

About 10 minutes.

I also have two smart plugs that can measure the current consumption and on average my computer with the i5-8500 GPU and laptop used about 100W (close to the max draw) when running Ollama.

In conclusion, if you're thinking of using gpt-oss:20b on (older) consumer-based computers, be prepared to wait for your output.