What is Parameters?

The internal values a model learns during training. More parameters can mean more capability, but also higher cost and slower inference.

Parameters are the numerical weights inside a neural network that get adjusted during training. They encode everything the model has learned, and their count is the standard shorthand for model size: a 7-billion-parameter model is small by current standards, while frontier models are believed to run into the hundreds of billions. More parameters generally mean more capacity for knowledge and reasoning, but also more memory, more compute per request, higher cost, and slower responses. The relationship is not linear, and training data quality, architecture, and post-training matter as much as raw size. Small models have improved fast and now handle many everyday tasks well. Note that mixture-of-experts models complicate the count: they may have a large total parameter count but activate only a fraction per request.

Example

Llama 3.1 was released in 8B, 70B, and 405B parameter sizes. The 8B version runs on a single consumer GPU; the 405B version needs a server cluster.

Why it matters

Parameter count is a quick proxy for a model's capability and running cost, useful when deciding between a cheap small model and an expensive large one. Browse the AI tools directory or the model leaderboard to put it into practice.