Active parameters
The number of a model’s parameters that are used to process each token. In a standard dense model every parameter is active. In a mixture-of-experts model only a fraction is, so the active count can be far smaller than the total.
Why it matters
Active parameters largely set the computing work per token, which drives response speed and running cost. Total parameters set how much memory the model needs, because every weight must be loaded even if only some are used for each token.
Comparing models on total size alone can mislead. Two models with similar total parameters can differ widely in speed and cost, and a smaller active count does not by itself say anything about quality. Capability still has to be measured on the organisation’s own tasks.
In practice
For example, a UK logistics firm choosing a self-hosted model for document extraction might shortlist a mixture-of-experts model because its low active count keeps responses fast, then size its servers on the much larger total parameter count.
Where Rodan fits
Rodan sizes models against measured task performance, latency and cost in Applied AI Engineering work. See also why falling AI costs make model portability a priority.