~/can-i-run/gpt-oss-20b
OpenAI

Can I run gpt-oss 20B?

Short answer: yes, on a M2 Max 64 (64GB) at FP16/BF16. Long answer below.

gpt-oss:20bLM StudiovLLMMLXoMLX

The math, in one paragraph.

$ ./vrambudget --explain gpt-oss-20b

gpt-oss 20B has 20.9B parameters (MoE: 3.6B active per forward pass, but all 20.9B must fit in memory). At FP16 that's 42 GB of raw weights. Quantization shrinks that, but you also need budget for the KV cache (definition), framework overhead, and safety headroom. The rule of thumb: real usable budget on a card is roughly its nameplate VRAM minus 25%. That's how the table below was computed.

What hardware actually fits.

$ grep "fits" gpus.json
FP16/BF16
42GB
17 GPUs fit
M2 Max 6464GBM3 Max 6464GBM4 Pro 6464GBM5 Pro 6464GB+ 13 more
Q8_0
22GB
22 GPUs fit
RTX 509032GBRTX A600048GBRTX 6000 Ada48GBL40S48GB+ 18 more
Q5_K_M
14GB
27 GPUs fit
RTX 309024GBRTX 3090 Ti24GBRTX 409024GBRX 7900 XTX24GB+ 23 more
Q4_K_M
12GB
33 GPUs fit
RTX 4060 Ti 16GB16GBRTX 4070 Ti Super16GBRTX 408016GBRTX 4080 Super16GB+ 29 more
Q3_K_M
9.0GB
33 GPUs fit
RTX 4060 Ti 16GB16GBRTX 4070 Ti Super16GBRTX 408016GBRTX 4080 Super16GB+ 29 more

Pick your path.

$ ls strategies/
Tightest budget

Smallest GPU that fits gpt-oss 20B at any quant: M2 Max 64 at FP16/BF16.

Reference quality (FP16)

Lossless inference needs 42 GB. Pick from 17 cards.

Best quality on a 24GB card

Q5_K_M fits comfortably (14 GB weights).

Tune the math yourself

Open the calculator pre-tuned for gpt-oss 20B: ↗ /calc?model=gpt-oss-20b

See the full model page.

$ ./open
$ opengpt-oss 20B20.9BOpenAI · ctx 128K$ rankBest GPU for gpt-oss 20Bsee the ranking →ranked by fit quality

Discussion.

$ gh discussion list

// sign in with github to leave a comment. threads live in the repo's discussions tab.