A learned value in a neural network that influences outputs; parameter count (often in billions) indicates model size. Parameter count is often cited as a capability proxy, but actual performance depends on architecture, training data, and post-training, not just size.
See: Hyperparameter; Model; Weights