Mac miniAvailable
Mac mini M5 Pro
15-core CPU / 16-core GPU, 24GB
$1,699at Apple Store
Cheapest that fits. Lowest price we've seen on any tracked rig that holds GPT-OSS 20B at 4-bit (13 of 17 GB usable).
Fits~82 tok/sMXFP4 (native)
Check priceOpenAI · 21B total / 3.6B active · MoE
Best hardware to run GPT-OSS 20B locally
GPT-OSS 20B needs about 13 GB at MXFP4 (native). 17 of the 17 priced rigs we track can hold it. Here's the cheapest, the best value and the fastest. Prices & stock checked 2026-10-23. Prices change daily and some retailers block our checker, so always confirm on the store page.
Our picks
Mac miniAvailable
Mac mini M5 Pro
15-core CPU / 16-core GPU, 24GB
$1,699at Apple Store
Best value. Lowest price per token/sec among complete systems where it fits comfortably: about $21 per tok/s at 4-bit. Same machine as the cheapest pick.
Fits~82 tok/sMXFP4 (native)
Check price
RTX 5090In stock
NVIDIA GeForce RTX 5090 Founders Edition
32GB card
$8,500at Newegg
Fastest. Most memory bandwidth of the rigs that fit it (1,792 GB/s), so the quickest decode: ~264 tok/s at 4-bit.
Fits~264 tok/sMXFP4 (native)
Suggested quant: Q8_0 (~197 tok/s)
GPU price is card only: you still need a host PC.
Check priceWhich quant to use
Use MXFP4 (native) on most rigs: it needs ~13 GB. Q8_0 (~25 GB) is a bit more accurate and fits comfortably on 14 of these rigs, e.g. the GMKtec EVO-X2. GPT-OSS 20B ships natively in MXFP4 (native), so that's the format to grab.
Every rig that fits it
| Machine | Price | 4-bit fit & speed | Suggested quant | $ per tok/s |
|---|---|---|---|---|
| Mac mini M5 ProMac mini | $1,699 | Fits~82 tok/s | MXFP4 (native) | $21 |
| GMKtec EVO-X2Strix Halo | $2,200 | Fits~69 tok/s | Q8_0 | $32 |
| Mac Studio M5 MaxMac Studio | $2,499 | Fits~101 tok/s | MXFP4 (native) | $25 |
| Mac Studio M5 MaxMac Studio | $3,099 | Fits~114 tok/s | Q8_0 | $27 |
| Framework Desktop DIYStrix Halo | $3,449 | Fits~69 tok/s | Q8_0 | $50 |
| GMKtec EVO-X2Strix Halo | $3,500 | Fits~69 tok/s | Q8_0 | $51 |
| Minisforum MS-S1 MaxStrix Halo | $3,919 | Fits~69 tok/s | Q8_0 | $57 |
| Beelink GTR9 ProStrix Halo | $4,349 | Fits~69 tok/s | Q8_0 | $63 |
| Gigabyte GeForce RTX 4090 Windforce V2RTX 4090 (card only, needs a host PC) | $4,899 | Fits~209 tok/s | MXFP4 (native) | $23 |
| Mac Studio M5 UltraMac Studio | $5,499 | Fits~119 tok/s | Q8_0 | $46 |
| Lenovo ThinkStation PGXDGX Spark / GB10 | $5,859 | Fits~80 tok/s | Q8_0 | $73 |
| ASUS Ascent GX10DGX Spark / GB10 | $5,999 | Fits~80 tok/s | Q8_0 | $75 |
| NVIDIA DGX SparkDGX Spark / GB10 | $6,950 | Fits~80 tok/s | Q8_0 | $87 |
| HP ZGX Nano G1nDGX Spark / GB10 | $7,699 | Fits~80 tok/s | Q8_0 | $96 |
| Gigabyte AI TOP ATOMDGX Spark / GB10 | $8,499 | Fits~80 tok/s | Q8_0 | $106 |
| NVIDIA GeForce RTX 5090 Founders EditionRTX 5090 (card only, needs a host PC) | $8,500 | Fits~264 tok/s | Q8_0 | $32 |
| PNY NVIDIA RTX PRO 6000 BlackwellRTX PRO 6000 (card only, needs a host PC) | $17,899 | Fits~264 tok/s | Q8_0 | $68 |
Sorted by lowest price. Speeds are estimated single-user decode at ~8K context. Full fit table, including rigs that don't fit: GPT-OSS 20B memory requirements. How we estimate
FAQ
How much memory does GPT-OSS 20B need?
About 13 GB at MXFP4 (native) and 25 GB at Q8_0, including runtime overhead and ~8K tokens of context for one user. Longer context needs more.
What's the cheapest hardware to run GPT-OSS 20B locally?
Of the rigs we track, the Mac mini M5 Pro at $1,699, estimated ~82 tok/s at 4-bit. Prices checked 2026-10-23.
Which quant should I use for GPT-OSS 20B?
MXFP4 (native) is the default: about 13 GB. Q8_0 needs about 25 GB; use it only if you have the room and still get usable speed.
How fast does GPT-OSS 20B run?
It has 3.6B active parameters per token, so speed is set by memory bandwidth: from ~69 tok/s on the slowest rig that fits to ~264 tok/s on the NVIDIA GeForce RTX 5090 Founders Edition. Estimates, not benchmarks.