GLM-5.2
A frontier-scale open-weights model. At 753 billion parameters it sits near the top of the open leaderboard, which is exactly why it needs the memory math below taken seriously.
What "parameters" means
A parameter is one small number the model adjusted while it was trained — roughly, one tiny piece of learned knowledge. GLM-5.2 has 753 billion of them. Every single one has to be stored somewhere the GPU can read it instantly while the model is running, which is why parameter count translates almost directly into how much GPU memory you need.
The memory math
Why the extra 20%? Running a model needs more than just the raw parameters sitting in memory — there's working room for the active computation, the prompt/response in flight, and normal overhead. Buying exactly the raw amount leaves no room and the model can crash or slow to a crawl under real use, so we always add 20% before recommending hardware.
Minimum hardware build that can run it
1× NVIDIA H200 SXM Server (8-GPU)
- Total GPU memory: 1128 GB (needs ≥ 904 GB — this build clears it)
- Total power draw: 10200 W — about 8.5 homes, 245 kWh/day, 2.7 EV batteries/day
- Total price: $380,000