Outmake Models
Models

GPT-OSS 20B

OpenAI · August 2025

GetHugging Face

The fastest chat model Outmake recommends for an 18 GB Mac, made for research and tool use.

Rating
3.2
Downloads
1M
Parameters
21B, 3.6B active
Context
128K
Download
11 to 14 GB
Commercial use
Yes

Tight fit on M4 · 16 GB

Outmake picks the UD-Q4_K_XL size: a 12 GB download that uses 12 GB of memory, and writes about 16 words (21 tokens) a second.

About

OpenAI meant it for local use and for replies that come back with little wait. OpenAI released its weights and tools openly so others can use and study it.

How it compares

Outmake estimates Qwen3.5 4B writes at about the same speed and can read pictures, which GPT-OSS 20B can't.

On Hugging Face it was 6th most downloaded last month of the 9 chat models Outmake recommends for an 18 GB Mac.

In LM Arena's public vote among open models, it places 43rd of 62 for text and 45th of 59 for coding.

Good to know

You can set how much it reasons before answering: low, medium or high.

It can be further trained for your own tasks on consumer hardware.

Sources: openai/gpt-oss-20b model card, gpt-oss-120b & gpt-oss-20b Model Card, gpt-oss-120b & gpt-oss-20b Model Card, Outmake's speed estimates, LM Arena and Hugging Face. Every line comes from one of these.

Ratings

3.2 out of 5, from how people voted on its answers.

Text
1317
Coding
1368

LM Arena ratings of October 28, 2026 (CC BY 4.0): people vote between two answers without knowing which model wrote which. The bar ends at 1488, the best any open model has. The Arena tests the full model; a small size on a Mac answers a little worse.

Good at

  • Everyday
  • Reasoning
  • Coding

Licence

Apache 2.0. Commercial use: yes. The maker's page has the full terms.

GPT-OSS 20B runs well from a Mac with an M3 Pro and 18 GB of memory, at UD-Q4_K_XL, writing about 19 words (26 tokens) a second. 48 of 59 Mac configurations run it well.

Compare with

The nearest models in size, judged for the same Mac.

Sizes

SizeDownloadMemory at 8KMemory at 32KOpen
Q3_K_MSmaller12 GB12 GB12 GBOpen
UD-Q4_K_XLBalanced12 GB12 GB12 GBOpen
UD-Q8_K_XLBest13 GB13 GB13 GBOpen
All 16 sizes
SizeDownloadMemory at 8KMemory at 32KOpen
Q3_K_S11 GB12 GBestimated12 GBOpen
Q2_K11 GB12 GBestimated12 GBOpen
Q4_012 GB12 GBestimated12 GBOpen
Q4_112 GB12 GBestimated12 GBOpen
Q4_K_S12 GB12 GBestimated12 GBOpen
Q4_K_M12 GB12 GBestimated12 GBOpen
Q5_K_S12 GB12 GBestimated12 GBOpen
Q5_K_M12 GB12 GBestimated12 GBOpen
Q2_K_L12 GB12 GBestimated12 GBOpen
Q6_K12 GB12 GBestimated12 GBOpen
UD-Q6_K_XL12 GB12 GBestimated12 GBOpen
Q8_012 GB12 GBestimated12 GBOpen
F1614 GB14 GBestimated14 GBOpen

Memory is what llama.cpp's own calculator reported for each size when the catalog was built, with the app's settings. Smaller, Balanced and Best are the three sizes the store names.

Run It Anywhere

GPT-OSS 20B is an open model, so any app that runs GGUF files can run it. These take the UD-Q4_K_XL size from unsloth straight from Hugging Face.

  • LM StudioMac, Windows, LinuxOpen
  • JanMac, Windows, LinuxOpen
  • MstyMac, Windows, LinuxOpen
  • OllamaMac, Windows, Linuxollama run hf.co/unsloth/gpt-oss-20b-GGUF:UD-Q4_K_XL
  • llama.cppMac, Windows, Linuxllama-server -hf unsloth/gpt-oss-20b-GGUF:UD-Q4_K_XL
  • Docker Model RunnerMac, Windows, Linuxdocker model run hf.co/unsloth/gpt-oss-20b-GGUF:UD-Q4_K_XL
On a PC Graphics Card
Video memoryLargest size that fitsDownloadMemory at 8K
6 GBNone worth runningneeds 12 GB
8 GBNone worth runningneeds 12 GB
10 GBNone worth runningneeds 12 GB
11 GBNone worth runningneeds 12 GB
12 GBQ4_K_M12 GB12 GB
16 GBF1614 GB14 GB
20 GBF1614 GB14 GB
24 GBF1614 GB14 GB
32 GBF1614 GB14 GB
48 GBF1614 GB14 GB

The largest size, of those that keep three bits a weight or more, whose memory at an 8K chat is under 92% of the card's; the rest is the desktop's. Estimated from the files, the same way as for a Mac; speed on a PC is not estimated yet. A size that doesn't fit the card still runs, slower, with part of it in ordinary memory.

On every Mac
MacSizeFitMemoryWritesReads
M6 · 16 GBUD-Q4_K_XLTight fit12 GB231223
M6 · 24 GBUD-Q8_K_XLRuns well14 GB211223
M6 · 32 GBUD-Q8_K_XLRuns great14 GB211223
M5 · 16 GBUD-Q4_K_XLTight fit12 GB21451
M5 · 24 GBUD-Q8_K_XLRuns well14 GB20451
M5 · 32 GBUD-Q8_K_XLRuns great14 GB20451
M5 Pro · 24 GBUD-Q8_K_XLRuns well14 GB38903
M5 Pro · 48 GBUD-Q8_K_XLRuns great14 GB38903
M5 Pro · 64 GBUD-Q8_K_XLRuns great14 GB381129
M5 Max · 36 GBUD-Q8_K_XLRuns great14 GB551806
M5 Max · 48 GBUD-Q8_K_XLRuns great14 GB702257
M5 Max · 64 GBUD-Q8_K_XLRuns great14 GB702257
M5 Max · 128 GBUD-Q8_K_XLRuns great14 GB702257
M5 Ultra · 96 GBUD-Q8_K_XLRuns great14 GB1233612
M5 Ultra · 256 GBUD-Q8_K_XLRuns great14 GB1234515
M5 Ultra · 512 GBUD-Q8_K_XLRuns great14 GB1234515
M4 · 16 GBUD-Q4_K_XLTight fit12 GB16152
M4 · 24 GBUD-Q8_K_XLRuns well14 GB15152
M4 · 32 GBUD-Q8_K_XLRuns great14 GB15152
M4 Pro · 24 GBUD-Q8_K_XLRuns well14 GB29303
M4 Pro · 48 GBUD-Q8_K_XLRuns great14 GB29303
M4 Pro · 64 GBUD-Q8_K_XLRuns great14 GB29303
M4 Max · 36 GBUD-Q8_K_XLRuns great14 GB39607
M4 Max · 48 GBUD-Q8_K_XLRuns great14 GB47758
M4 Max · 64 GBUD-Q8_K_XLRuns great14 GB47758
M4 Max · 96 GBUD-Q8_K_XLRuns great14 GB39607
M4 Max · 128 GBUD-Q8_K_XLRuns great14 GB47758
M3 · 8 GBQ3_K_MToo big13 GB
M3 · 16 GBUD-Q4_K_XLTight fit12 GB14126
M3 · 24 GBUD-Q8_K_XLRuns well14 GB13126
M3 Pro · 18 GBUD-Q4_K_XLRuns well12 GB19220
M3 Pro · 36 GBUD-Q8_K_XLRuns great14 GB18220
M3 Max · 36 GBUD-Q8_K_XLRuns great14 GB32471
M3 Max · 48 GBUD-Q8_K_XLRuns great14 GB40628
M3 Max · 64 GBUD-Q8_K_XLRuns great14 GB40628
M3 Max · 96 GBUD-Q8_K_XLRuns great14 GB32471
M3 Max · 128 GBUD-Q8_K_XLRuns great14 GB40628
M3 Ultra · 96 GBUD-Q8_K_XLRuns great14 GB63942
M3 Ultra · 256 GBUD-Q8_K_XLRuns great14 GB63942
M3 Ultra · 512 GBUD-Q8_K_XLRuns great14 GB631256
M2 · 8 GBQ3_K_MToo big13 GB
M2 · 16 GBUD-Q4_K_XLTight fit12 GB14123
M2 · 24 GBUD-Q8_K_XLRuns well14 GB13123
M2 Pro · 16 GBUD-Q4_K_XLTight fit12 GB25247
M2 Pro · 32 GBUD-Q8_K_XLRuns great14 GB23247
M2 Max · 32 GBUD-Q8_K_XLRuns great14 GB40462
M2 Max · 64 GBUD-Q8_K_XLRuns great14 GB40462
M2 Max · 96 GBUD-Q8_K_XLRuns great14 GB40586
M2 Ultra · 64 GBUD-Q8_K_XLRuns great14 GB62925
M2 Ultra · 128 GBUD-Q8_K_XLRuns great14 GB62925
M2 Ultra · 192 GBUD-Q8_K_XLRuns great14 GB62925
M1 · 8 GBQ3_K_MToo big13 GB
M1 · 16 GBUD-Q4_K_XLTight fit12 GB9.1100
M1 Pro · 16 GBUD-Q4_K_XLTight fit12 GB22200
M1 Pro · 32 GBUD-Q8_K_XLRuns great14 GB21200
M1 Max · 32 GBUD-Q8_K_XLRuns great14 GB34343
M1 Max · 64 GBUD-Q8_K_XLRuns great14 GB34343
M1 Ultra · 64 GBUD-Q8_K_XLRuns great14 GB49685
M1 Ultra · 128 GBUD-Q8_K_XLRuns great14 GB49685

Writes and reads are words a second (a token is about three quarters of a word), 4K tokens into a chat, estimated unless marked measured. The size is the one the store offers that Mac. Constants fitted October 11, 2026; how, and how wrong.

Badge for the model card

Shows on GPT-OSS 20B's Hugging Face card which Macs run it well, linked to this page. It updates every morning.

Runs on Mac badge

Sizes and files are read from unsloth/gpt-oss-20b-GGUF on Hugging Face and checked every morning. The rating is the model's score on the LM Arena text leaderboard (CC BY 4.0), where people vote between two answers without knowing which model wrote which. Catalog of November 4, 2026.

Your Computer

Every model is judged for this Mac. Both are under About This Mac, in the Apple menu.

Every Mac · Every graphics card

Civitai

Open in OutmakeHugging Face

New to Outmake? Download the free beta