AI glossary
Parameters
The internal numbers (mostly weights) a model learns during training. Their count is a rough measure of its size.
A neural network is a huge function, and its parameters are the values that define it. Training adjusts them, one tiny step at a time, until the model’s outputs improve. Today’s large models have hundreds of billions of parameters or more.
More parameters usually means more capacity, but not automatically a better model: data quality, training method and architecture matter just as much. Many companies no longer publish parameter counts for their closed models.
Example: “552B” in a model description means 552 billion parameters.
In practice
- The parameter count hints at size, but compare models with real tests, not by the number.
- To run a model locally, the parameter count tells you roughly how much memory you will need.
- With mixture-of-experts models, also look at the active parameters.