← Back
Startup path

Which model do you want to run?

Pick one — we'll show the cheapest real cluster that can actually hold it.

MIT

GLM 5.2 Max

744B parameters · Z.AI

MIT

DeepSeek V4 Pro

1,600B parameters · DeepSeek

OpenMDW

Nemotron 3 Ultra

550B parameters · NVIDIA

Recommended build for GLM 5.2 Max

744B × 1 GB × 1.2 = 892.8 GB minimum

12x NVIDIA H100 SXM Cluster of 12 — the cheapest combination that clears the memory minimum.

Combined GPU memory960 GB
Combined power8,400 W (7.0x an average home)
Total price$384,000
Request this build See full math →