What Are Parameters? Why "Billions of Parameters" Matters
Model announcements always mention a parameter count — here's what that number actually represents and why it's used as a rough proxy for capability.
What a parameter actually is
A parameter is a single adjustable number inside a neural network — one weight on one connection between units (see What Is a Neural Network?). A model described as having "70 billion parameters" has that many individual numbers, all tuned during training.
Why the count gets quoted so often
More parameters generally mean more capacity to represent complex patterns, which is why parameter count became a shorthand for "how capable is this model." It's an imperfect proxy, though — training data quality and methodology matter enormously too, and a smaller, well-trained model can outperform a larger, poorly-trained one on real tasks.
The practical tradeoff
More parameters also means more compute needed to run the model, higher cost per request, and slower responses — which is exactly why providers ship a range of model sizes (see Choosing the Right Model Size) rather than just one giant model for every use case.
Frequently Asked Questions
What does "parameters" mean when talking about AI models?
The individual adjustable numbers inside a neural network — the weights on its internal connections — that get tuned during training. A model's parameter count is often quoted as a rough (imperfect) proxy for its capability.
Does more parameters always mean a better model?
Not necessarily — training data quality and methodology matter enormously. A smaller model trained well can outperform a larger model trained poorly on real-world tasks.