FitLocally by GPUwerk

OpenAI · 21B · MoE

Can you run gpt-oss-20b locally?

gpt-oss-20b is a mixture-of-experts model with 21B parameters, 3.6B active per token. At 4-bit or better it runs on 85 of the 120 hardware configurations listed here; the smallest is the RX 7900 XT (20 GB). On a 128 GB DGX Spark it runs at MXFP4, at roughly 63 tokens per second (our est.).

Data as of 2 November 2026

Fastest machines for gpt-oss-20b

Every machine that runs it, fastest first. Filter by type; your machine is marked. Compare machines running it side by side

  1. 1RTX 5090Graphics card · 32 GB~261 tok/s ModelFit est.MXFP4 · 15.1 of 31 GB
  2. 2RTX PRO 6000 BlackwellGraphics card · 96 GB~261 tok/s ModelFit est.MXFP4 · 15.1 of 95 GB
  3. 32x RTX 5090Graphics card · 64 GB~242 tok/s ModelFit est.MXFP4 · 15.1 of 62 GB
  4. 42x RTX PRO 6000 BlackwellGraphics card · 192 GB~242 tok/s ModelFit est.MXFP4 · 15.1 of 190 GB
  5. 5RTX 4090Graphics card · 24 GB~169 tok/s ModelFit est.MXFP4 · 15.1 of 23 GB
  6. 6M5 UltraMac · 96–512 GB · 5 variants~167 tok/s ModelFit est.MXFP4 · 15.1 GB
  7. 72x RTX 4090Graphics card · 48 GB~161 tok/s ModelFit est.MXFP4 · 15.1 of 46 GB
  8. 8RTX 3090 TiGraphics card · 24 GB~157 tok/s ModelFit est.MXFP4 · 15.1 of 23 GB
  9. 92x RTX 3090 TiGraphics card · 48 GB~150 tok/s ModelFit est.MXFP4 · 15.1 of 46 GB
  10. 10RTX 3090Graphics card · 24 GB~148 tok/s ModelFit est.MXFP4 · 15.1 of 23 GB
  11. 112x RTX 3090Graphics card · 48 GB~142 tok/s ModelFit est.MXFP4 · 15.1 of 46 GB
  12. 12RX 7900 XTXGraphics card · 24 GB~126 tok/s ModelFit est.MXFP4 · 15.1 of 23 GB
  13. 132x RX 7900 XTXGraphics card · 48 GB~122 tok/s ModelFit est.MXFP4 · 15.1 of 46 GB
  14. 14M3 UltraMac · 96–512 GB · 5 variants~117 tok/s ModelFit est.MXFP4 · 15.1 GB
  15. 15RX 7900 XTGraphics card · 20 GB~112 tok/s ModelFit est.MXFP4 · 15.1 of 19.0 GB
  16. 16M5 MaxMac · 48–128 GB · 3 variants~109 tok/s ModelFit est.MXFP4 · 15.1 GB
  17. 17M2 UltraMac · 64–192 GB · 7 variants~103 tok/s ModelFit est.MXFP4 · 15.1 GB
  18. 18M1 UltraMac · 64–128 GB · 4 variants~92 tok/s ModelFit est.MXFP4 · 15.1 GB
  19. 19M4 MaxMac · 48–128 GB · 3 variants~92 tok/s ModelFit est.MXFP4 · 15.1 GB
  20. 20M5 Max · 36 GB · 32cMac · 36 GB~85 tok/s ModelFit est.MXFP4 · 15.1 of 24 GB
  21. 21M3 MaxMac · 48–128 GB · 3 variants~72 tok/s ModelFit est.MXFP4 · 15.1 GB
  22. 22M4 Max · 36 GB · 32cMac · 36 GB~71 tok/s ModelFit est.MXFP4 · 15.1 of 24 GB
  23. 23M2 MaxMac · 32–96 GB · 5 variants~67 tok/s ModelFit est.MXFP4 · 15.1 GB
  24. 24M5 ProMac · 24–64 GB · 6 variants~64 tok/s ModelFit est.MXFP4 · 15.1 GB
  25. 25DGX Spark (GB10)AI mini PC · 128 GB~63 tok/s our est.MXFP4 · 15.1 of 122 GB
  26. 26M1 MaxMac · 32–64 GB · 4 variants~60 tok/s ModelFit est.MXFP4 · 15.1 GB
  27. 27Ryzen AI Max+ 395 · 64 GBAI mini PC · 64 GB~60 tok/s our est.MXFP4 · 15.1 of 48 GB
  28. 28M3 MaxMac · 36–96 GB · 2 variants~55 tok/s ModelFit est.MXFP4 · 15.1 GB
  29. 29M4 ProMac · 24–64 GB · 6 variants~53 tok/s ModelFit est.MXFP4 · 15.1 GB
  30. 30Ryzen AI Max+ 395 · 128 GBAI mini PC · 128 GB~42 tok/s ModelFit est.MXFP4 · 15.1 of 96 GB
  31. 31M2 Pro · 32 GBMac · 32 GB~39 tok/s ModelFit est.MXFP4 · 15.1 of 21 GB
  32. 32M6Mac · 24–32 GB · 3 variants~38 tok/s ModelFit est.MXFP4 · 15.1 GB
  33. 33M1 Pro · 32 GBMac · 32 GB~35 tok/s ModelFit est.MXFP4 · 15.1 of 21 GB
  34. 34M5Mac · 24–32 GB · 4 variants~32 tok/s ModelFit est.MXFP4 · 15.1 GB
  35. 35M3 Pro · 36 GBMac · 36 GB~31 tok/s ModelFit est.MXFP4 · 15.1 of 24 GB
  36. 36M4Mac · 24–32 GB · 2 variants~24 tok/s ModelFit est.MXFP4 · 15.1 GB
  37. 37M3 · 24 GBMac · 24 GB~21 tok/s ModelFit est.MXFP4 · 15.1 of 16.0 GB
  38. 38M2 · 24 GBMac · 24 GB~20 tok/s ModelFit est.MXFP4 · 15.1 of 16.0 GB
Too big for 35 machines

RTX 5080, RTX 5070 Ti, RTX 5060 Ti, RTX 4080 SUPER, RTX 4080, RTX 4070 Ti SUPER, RTX 4060 Ti, Tesla T4, Tesla P100, RX 9070 XT, RX 9060 XT, M3 Pro · 18 GB, RTX 5070, RTX 4070 Ti, RTX 4070 SUPER, RTX 4070, RTX 3060, M2 Pro · 16 GB, M1 Pro · 16 GB, M6 · 16 GB · 12c, M6 · 16 GB · 10c, M5 · 16 GB · 10c, M5 · 16 GB · 8c, M4 · 16 GB, M3 · 16 GB, M2 · 16 GB, M1 · 16 GB, RTX 5060, RTX 4060, RTX 3070, RTX 3060 Ti, RTX 2070, M3 · 8 GB, M2 · 8 GB, M1 · 8 GB. gpt-oss-20b needs about 15.1 GB at its smallest build.

Speeds are single-user decode estimates, not measurements. "ModelFit est." comes from ModelFit's model fitted on published runs (CC BY 4.0); "our est." is memory bandwidth divided by bytes read per token, times 0.55. Memory is what the best build needs, against what the machine can give a model. Method

DGX Spark setup notes: GPUwerk guide for gpt-oss-20b.

About the model

Parameters
21B
Active per token
3.6B
Type
Mixture of experts
Good for
General chat on modest hardware
Model card
Hugging Face
Hugging Face count
20.91B (agrees)

Builds and how to run them

QuantizationLoadMin. RAMCommand
MXFP413.8 GB24 GBollama run gpt-oss:20b

Load, minimum RAM and Ollama commands: ModelFit, CC BY 4.0. Load is memory with the model loaded, before long contexts.

Questions

Can I run gpt-oss-20b on a Mac?

Yes on a Mac with enough memory: the smallest Apple configuration listed here that runs it at 4-bit or better is the M5 Pro · 24 GB · 20c.

How much memory does gpt-oss-20b need?

About 15.1 GB at the smallest build in our data, including a margin for the runtime and an 8K-token context. Longer contexts need more.

Does gpt-oss-20b run on a DGX Spark?

Yes, at MXFP4. The Spark has 128 GB of unified memory, about 121.6 GB usable.