GLM-5.2 (Max)
Currently the highest-ranked MIT-licensed model on the public agent leaderboard — built for multi-step, tool-using AI agents, not just chat.
📊 Arena.ai Agent Leaderboard, rank 9 — the strongest fully open-license model we could confirm on the agent leaderboard. · source
The memory math
parameters = 753 billion
× 1 GB per billion = 753 GB (raw weight size)
× 1.2 (20% working room for the model to actually run, not just sit in storage)
= 903.6 GB minimum memory needed
× 1 GB per billion = 753 GB (raw weight size)
× 1.2 (20% working room for the model to actually run, not just sit in storage)
= 903.6 GB minimum memory needed
Why the 20% headroom? A model doesn't just sit in memory — while it's answering you, it needs extra scratch space to hold the conversation so far and its in-progress calculations. Buy exactly the raw size and it will crash under real use.
Two ways to get there
Option A
10× NVIDIA RTX PRO 6000 Blackwell Server Edition$130,000Total memory960 GB
Total power6.0 kW
Total price$130,000
10 cards × 96 GB = 960 GB, comfortably over the 903.6 GB needed. Cheaper than a DGX B200, but 10 independently-networked cards are harder to keep fed at full speed than one integrated system.
Request this Quote✅ Recommended: Option B
1× NVIDIA DGX B200$515,410Total memory1,440 GB
Total power14.3 kW
Total price$515,410
One DGX B200 (1,440 GB, real NVLink cluster) fits this model with 536 GB to spare and is what a production deployment would actually use.
Request this QuoteWhat that power means
🏠
Running the recommended build 24/7 draws roughly 11.9 average homes' worth of continuous power.
🔋
One hour of running it uses about 15.9% of a typical 90 kWh electric-car battery.