Definition
Parameters are the internal numbers an AI model tunes during training to make accurate predictions, and weights are the main type, controlling how strongly each input influences the result.[1]
At a glance
- A parameter is just a number; weights are the most common kind, setting how much one piece of input matters[2].
- Training is the process of adjusting these numbers until the model’s answers get reliably better[1].
- Bigger models have more parameters (GPT-3 ~175 billion, GPT-4 estimated ~1.8 trillion), which usually means more capability but higher running cost[3].
- The full set of trained parameters IS the model; sharing those numbers is what people mean by open-weight models.
The recipe analogy
Think of a cookie recipe: 2 cups flour, 1 cup sugar. Those numbers control the outcome; change them and you get different cookies. Parameters work the same way, except an AI has billions of them and learns the right values automatically by tasting its own results millions of times[4].
Why the count matters to you
Parameter count is a rough proxy for how much a model knows and can do. More parameters often means smarter output, but also more computing power, slower responses, and higher cost per use. A smaller, cheaper model is frequently the better business choice for routine tasks[3].
Bottom line
Parameters and weights are the learned numbers that make an AI work; their count signals capability but also cost, so bigger is not always better for your needs.