Can I run it?

Pick a model. We'll show you the cheapest hardware that runs it, the box we'd actually recommend, and the fastest option, with real token per second estimates.

Which model do you want to run?

GLM 4.6 355B-A32B (MoE)•355B•GLM

GeneralCodingReasoningMathCreative

Runs on 7 builds

RecommendedFastest

NVIDIA DGX Station (GB300, 784GB)

DGX

$80,000

Q4_K_M•medium

210.0 GB784 GB

Good•24.3 t/s

Buy on NVIDIA · $80,000

Minimum

Mac Studio M3 Ultra (256GB)

Apple Silicon

$7,499

Q4_K_M•medium

210.0 GB256 GB

Very Slow•2.8 t/s

Buy on Apple · $7,499

Every build that runs GLM 4.6 355B-A32B (MoE)

Mac Studio M3 Ultra (256GB)

Apple Silicon · Q4_K_M

2.8 t/s$7,499

Mac Studio M3 Ultra (512GB)

Apple Silicon · Q4_K_M

2.8 t/s$9,499

4x DGX Spark Cluster (512GB)

DGX · Q4_K_M

0.9 t/s$16,000

2x Mac Studio M3 Ultra (1TB cluster)

Apple Silicon · Q4_K_M

2.8 t/s$19,000

8x RTX PRO 6000 Blackwell (768GB)

Multi GPU · Q4_K_M

6.1 t/s$78,000

NVIDIA DGX Station (GB300, 784GB)

DGX · Q4_K_M

24.3 t/s$80,000

8x NVIDIA H200 SXM (1.1TB)

Multi GPU · Q4_K_M

16.2 t/s$300,000

Token per second figures are estimates derived from memory bandwidth, not benchmarks. "Recommended" means the cheapest build that runs this model at good quality and usable speed. Some links are affiliate links.