A parameter is a value an AI model learned in training, and the parameter count is the usual measure of a model's size. Models are often described by the number of billions of parameters they contain, and parameters are what the weights are made of.
It matters because the count gives a rough idea of how much memory a model needs and how heavy it is to run. A model with more parameters generally needs more memory and more computation, although the relationship with quality is not simple.
The number alone is not a measure of ability. Well-trained smaller models can match larger ones on some tasks, and techniques such as mixture-of-experts mean that only some parameters are used at a time, so the total count and the active count may differ. The memory needed also depends on the precision at which parameters are stored.
When comparing models, look at both the parameter count and the precision, check the size in memory, and judge quality using tests on tasks that resemble your use.