A. Smyntyna / Code
ENRU

active parameters

How much of a model actually runs to produce one token, as opposed to how much of it sits in memory. In a name like qwen3.6-35b-a3b the a3b is this number: 35 billion parameters stored, 3 billion of them used per token. It predicts speed, where the total predicts memory.

A dense model has no such number, because the answer is always all of them. The a-number only appears on a mixture of experts, and its absence is itself information: no a-number means dense.

The two numbers pull in opposite directions when you shop. Total parameters decide whether the file fits. Active parameters decide whether waiting for it is bearable.

Read next