Compare machines for local AI
Pick a model and up to three machines: see how fast each one runs it, then memory, bandwidth and what else fits, row by row.
Data as of 2 November 2026. Speeds are estimates unless marked measured.
Showing two of three machines. Turn your phone sideways for all three.
Running gpt-oss-20b
Runs
~63 tok/s
MXFP4 · 15.1 of 122 GB · our est.
Runs
~92 tok/s
MXFP4 · 15.1 of 96 GB · ModelFit est.
Runs
~261 tok/s
MXFP4 · 15.1 of 31 GB · ModelFit est.
Single-user decode speed, estimated. Fastest machines for this model
Memory
128 GB
memory
128 GB
memory
32 GB
memory
~122 GB
usable for a model
~96 GB
usable for a model
~31 GB
usable for a model
273 GB/s
memory bandwidth
546 GB/s
memory bandwidth
1792 GB/s
memory bandwidth
What it runs
103
of 110 models run at 4-bit or better
101
of 110 models run at 4-bit or better
86
of 110 models run at 4-bit or better
Our pick
Q4_K_XL · ~28 tok/s our est.
Nemotron 3.5 Super VL 120B-A12B
Q4_K_M · ~41 tok/s our est.
Q4_K_M · ~346 tok/s ModelFit est.
Largest current model that fits
125B at Q4_K_XL
120B at Q4_K_M
35B at Q4_K_M
Popular models
Runs
MXFP4 · ~61 tok/s
Runs
MXFP4 · ~77 tok/s
Too big
needs ~68 GB
Runs
MXFP4 · ~63 tok/s
Runs
MXFP4 · ~92 tok/s
Runs
MXFP4 · ~261 tok/s
Runs
Q8_0 · ~45 tok/s
Runs
Q8_0 · ~87 tok/s
Runs
Q4_K_M · ~346 tok/s
Runs
Q4_K_M · ~11 tok/s
Runs
Q8_0 · ~15 tok/s
Runs
Q8_0 · ~36 tok/s
Runs
Q8_0 · ~52 tok/s
Runs
Q8_0 · ~63 tok/s
Runs
Q8_0 · ~195 tok/s
Runs
Q4_K_M · ~6.0 tok/s slow
Runs
Q4_K_M · ~9.7 tok/s slow
Too big
needs ~44 GB
Runs
Q4_K_M · ~23 tok/s
Runs
Q4_K_M · ~28 tok/s
Too big
needs ~67 GB
2-bit only
Q2_K · ~13 tok/s
Too big
needs ~177 GB
Too big
needs ~177 GB
Rent one by the hour
Use Change under a machine to pick another type, producer, model or variant. How the numbers are worked out