Parameters
The learned numerical values inside a model — the \"7 billion parameters\" figure vendors cite as a size signal, with bigger meaning more capable but also more expensive.
Parameters are the numerical values a model learns during training — billions of tiny adjustable weights that collectively encode everything the model has learned about patterns in its training data. When a vendor describes a model as having "7 billion" or "70 billion parameters," they're describing the model's scale. Larger parameter counts generally enable more complex reasoning and broader capabilities, but they also require significantly more compute to run, which translates directly to inference cost and latency.
Parameter count has become a marketing metric in ways that can mislead procurement decisions. A 70-billion parameter model costs substantially more per request than a 7-billion parameter model, but for many specific business tasks — classifying support tickets, extracting structured data from documents, summarizing a standard report — the smaller model performs comparably or better because it's been tuned for the task. Organizations that default to the largest available model without matching it to the task often overpay for capability they don't use, while also building on infrastructure that's slower and more expensive to scale.
Continue path
Latency
How response speed affects AI usability in production
Optional map
Concept neighborhood
Focused neighborhood
Parameters
The learned numerical values inside a model — the \"7 billion parameters\" figure vendors cite as a size signal, with bigger meaning more capable but also more expensive.
In these paths
Selected concept
Directly related
One step further
via Large Language Models
via Foundation Models
via Training Data