Parameters are the total count of learnable values (weights and biases) in a model, commonly used as a rough proxy for model capacity — '7B parameters' means 7 billion learned numbers. Parameter count influences a model's ability to memorize and generalize, but more parameters don't automatically mean better performance; data quality, architecture, and training methodology matter as well. Scaling laws describe predictable relationships between parameter count, training data, compute budget, and final performance.