AI glossary · Basic · Language models · Efficiency & infra
Active parameters
Active parameters are the number of parameters a model actually uses in computation when processing a single token. MoE models select and use only a portion of their total parameters, so the total and active parameter counts differ greatly.
About this term
- Level
- Basic · Technical terms you often meet in AI news
- Category
- Infrastructure & using models
- Added
- 2026-10-12
How it works
A standard model uses all of its parameters in computation for every token. An MoE model has several groups of "experts," and a router selects only some of the experts for each token to compute with. The parameters that actually take part in computation are the active parameters. This lets the model have the knowledge capacity of a large model while reducing the computation per token to the level of a small model.
How to get a feel for it
Model names or model cards often list both total and active parameters. For example, a name like "30B-A3B" is read as 30 billion total and 3 billion active parameters. Plan memory based on the total, and estimate weight size as "total number of parameters × bytes per number." Computation per token and generation speed, on the other hand, track the active parameters more closely.
Total parameters vs. active parameters
Total parametersAll values the model has
- Determines required memory
- All experts are loaded in memory
Active parametersValues actually used per token
- Determines computation per token
- Closer to generation speed
공통 In standard models the two numbers are the same; they differ only in MoE
Things to keep in mind
A small active parameter count does not mean the model fits on a small GPU; experts that are not used must still be loaded in memory. Quality is usually better than a standard model of the same active size, but it is not guaranteed to match a standard model of the same total size. When serving across multiple GPUs, communication between experts can become a bottleneck. When comparing models, always check whether a number refers to total or active parameters.
Example
Suppose a team that wants to run an open-weight model on in-house servers is reviewing an MoE model labeled something like "30B-A3B" (as an example). The person in charge checks whether there is enough VRAM based on the 30 billion total parameters for memory, and estimates throughput by expecting response speed to be close to that of a 3-billion-parameter model based on the active parameters. Actual speed varies with the inference engine and GPU configuration, so they confirm it by measuring directly.