M4 Max (128 GB, 40-core GPU) vs M3 Ultra (256 GB, 80-core GPU)
The M4 Max (128 GB, 40-core GPU) runs 101 of 110 open models at 4-bit or better, the M3 Ultra (256 GB, 80-core GPU) runs 105. Usable memory: about 96 GB against 192 GB. The M3 Ultra (256 GB, 80-core GPU) has more memory bandwidth (819 vs 546 GB/s), so models that fit both answer faster on it.
Data as of 2 November 2026. Speeds are estimates unless marked measured.
Running gpt-oss-120b
Runs
~77 tok/s
MXFP4 · 68 of 96 GB · ModelFit est.
Runs
~101 tok/s
MXFP4 · 68 of 192 GB · ModelFit est.
Single-user decode speed, estimated. Fastest machines for this model
Memory
128 GB
memory
256 GB
memory
~96 GB
usable for a model
~192 GB
usable for a model
546 GB/s
memory bandwidth
819 GB/s
memory bandwidth
What it runs
101
of 110 models run at 4-bit or better
105
of 110 models run at 4-bit or better
Our pick
Nemotron 3.5 Super VL 120B-A12B
Q4_K_M · ~41 tok/s our est.
Q4_K_M · ~57 tok/s our est.
Largest current model that fits
120B at Q4_K_M
284B at Q4_K_M
Popular models
Runs
MXFP4 · ~77 tok/s
Runs
MXFP4 · ~101 tok/s
Runs
MXFP4 · ~92 tok/s
Runs
MXFP4 · ~117 tok/s
Runs
Q8_0 · ~87 tok/s
Runs
Q8_0 · ~112 tok/s
Runs
Q8_0 · ~15 tok/s
Runs
Q8_0 · ~22 tok/s
Runs
Q8_0 · ~63 tok/s
Runs
Q8_0 · ~85 tok/s
Runs
Q4_K_M · ~9.7 tok/s slow
Runs
Q6_K · ~11 tok/s
Runs
Q4_K_M · ~28 tok/s
Runs
Q4_K_M · ~41 tok/s
Too big
needs ~177 GB
Runs
Q4_K_M · ~57 tok/s
Rent one by the hour
—
—
Use Change under a machine to pick another type, producer, model or variant. How the numbers are worked out