The hardware that actually runs enterprise AI.
Three MIT-licensed models. Four hardware tiers. Verified specs, honest math — we tell you what works and what doesn't.
DeepSeek V4 Flash
Smallest footprint, lowest build cost. Best starting point if you're not sure how much scale you need.
284B params · MoE
Needs 341 GB GPU memory
See builds →GLM 5.2
Mid-range scale. Significantly larger than Flash, roughly 2–3× the hardware cost.
753B params · MoE
Needs 904 GB GPU memory
See builds →DeepSeek V4 Pro
Maximum scale — the largest openly licensed model available. Highest hardware requirement and cost.
1.6T params · MoE
Needs 1920 GB GPU memory
See builds →Not sure which model or hardware you need?
Help me choose →