Model size
model sizes
How big a model is, measured by its number of parameters. Providers often offer whole families in several sizes: small and fast, medium all-rounders, large and especially capable.
💭 Imagine …
Like vehicles: an e-scooter is faster and cheaper around town, but for moving house you need the truck.
🔎 Example
Small models with a few billion parameters even run on a laptop or smartphone. The largest models have hundreds of billions of parameters or more and only run in data centres. Many providers don't publish the exact size.
🎯 Why it matters
Bigger isn't always better: for classifying, summarising or extracting, a small model is often enough, and it's faster and cheaper. Large models pay off for complex, open-ended tasks. Small models are often “distilled”, i.e. taught by larger ones.
🎓 Prefer step by step?
The AI Literacy course: 6 short modules with quizzes, a solid foundation in about 60 minutes.