Skip to content
selfhostr

run a model

Running gpt-oss 20B

OpenAI's open-weight model, in the size meant for one machine.

20 billion parameterslicence : Apache 2.0weights

memory needed

13.6 GB

the smallest hardware that runs it properly

Le moins cher à ce palier : Mac mini M4, 24 Go.

€1,169 · checked 2026-08-02

the alternatives

on the processor alone: slow, and it works

price not checked
price not checked
price not checked
€1,839 · checked 2026-08-02product page
€2,559 · checked 2026-08-02product page

on the processor alone: slow, and it works

price not checked
€4,099 · checked 2026-08-02product page
€7,659 · checked 2026-08-02product page

runs it, but only just

price not checked

runs it, but only just

price not checked

Only just means it works, until the day you lengthen the context or open something else. Which is the day you conclude the tool lied to you.

not enough memory

if you change the format

4-bit weightswhat almost everyone runs, and what the hardware below is judged against
13.6 GB
8-bit weightsvisibly better on long or precise answers
24.4 GB
16-bit weightsthe model as it was trained, and rarely worth it at home
48.4 GB

Everything above assumes 4-bit weights and a 8k token context. Going up in precision costs memory; shortening the context is the other lever, and the cheaper one when something almost fits. Model details checked on 2026-08-02.

how to actually run it

what this page will not tell you

How many tokens per second you will get. That is a benchmark — it depends on memory bandwidth, on the engine, on the context and on how long the answers are. This site does not invent benchmarks, and everyone who quotes one for your exact setup is guessing.

builder

Price a whole machine around itA card is not a server. The builder adds the machine, the drives, the electricity and the backup.