A model parameter is a numerical value learned during training that helps determine how a model behaves. Training produces the parameter values; using the model to analyze data puts those learned values to work. A provider may keep them private or release them with the model. section 8.4
Choosing a model for review classification
Suppose you want to classify customer reviews as positive or negative. You could send the reviews to a hosted model, whose learned parameter values remain private. Or you could choose a released model whose parameters are available and run it on your own infrastructure. The choice affects where your data goes, as well as what computing resources you need. section 8.4 section 1.5.2
Do not confuse learned model parameters with the settings you pass to an API request. In `chat.completions.create`, `model` selects which model to use, while other configuration parameters can influence its reply—for example, by limiting output length. Fine-tuning is different again: it trains an existing model on task-specific examples to create a specialized variant. section 3.4 section 1.5.2
