The Reckoning

Three great open-source models from the world's agent leaderboard (arena.ai/leaderboard/agent), filtered to Open Source — all under the MIT license, all well past 100 billion parameters. For each, the arithmetic that determines the smallest honest engine able to run it, shown in full.

MIT license — open source

DeepSeek V4 Flash

The smallest of our three flagship models — MoE keeps inference cheap (only 13B parameters active per token) but every expert still has to sit in GPU memory.

Total parameters 284B (Mixture-of-Experts (MoE), 13B active per token)

What it means: parameter count is the model's size. “284B” means 284 billion weights. Even for a Mixture-of-Experts model where only 13B are active per token, every expert must still reside in memory at once — one cannot decline to load the experts not needed this particular token.

The Reckoning of Memory = parameters (in billions) × 1 GB, plus 20% working room = 284 × 1 GB × 1.2 = 340.8 GB, minimum
Smallest engine that clears it H200 HGX 4-GPU NVLink Server (564 GB pooled)

564 GB ≥ 340.8 GB required, making this the smallest true configuration in our house that fits the model. Real cluster: these 4 GPUs share one NVLink memory domain, so their memory truly adds up to 564 GB.

Total price for this engine $140,000
Inquire & Purchase →
MIT license — open source

GLM-5.2 (Max)

A much larger MoE model — needs the full 8-GPU NVLink node to have any room to spare.

Total parameters 744B (Mixture-of-Experts (MoE), 40B active per token)

What it means: parameter count is the model's size. “744B” means 744 billion weights. Even for a Mixture-of-Experts model where only 40B are active per token, every expert must still reside in memory at once — one cannot decline to load the experts not needed this particular token.

The Reckoning of Memory = parameters (in billions) × 1 GB, plus 20% working room = 744 × 1 GB × 1.2 = 892.8 GB, minimum
Smallest engine that clears it H200 HGX 8-GPU NVLink Server (1128 GB pooled)

1128 GB ≥ 892.8 GB required, making this the smallest true configuration in our house that fits the model. A full 8-GPU NVSwitch domain — all 8 GPUs' memory is addressable as one 1,128 GB pool.

Total price for this engine $280,000
Inquire & Purchase →
MIT license — open source

DeepSeek V4 Pro

1.6 trillion parameters. Even a full 8-GPU H200 node (1,128 GB) can't hold it — this is a rack-scale-only model.

Total parameters 1600B (Mixture-of-Experts (MoE), 49B active per token)

What it means: parameter count is the model's size. “1600B” means 1600 billion weights. Even for a Mixture-of-Experts model where only 49B are active per token, every expert must still reside in memory at once — one cannot decline to load the experts not needed this particular token.

The Reckoning of Memory = parameters (in billions) × 1 GB, plus 20% working room = 1600 × 1 GB × 1.2 = 1920.0 GB, minimum
Smallest engine that clears it NVIDIA GB300 NVL72 (full rack) (20480 GB pooled)

20480 GB ≥ 1920.0 GB required, making this the smallest true configuration in our house that fits the model. The largest real cluster that exists today: 72 Blackwell Ultra GPUs across the rack share one 20 TB NVLink memory pool.

Total price for this engine $5,000,000
Inquire & Purchase →