Model size
Also known as: parameter count · 7B · 70B
Model size is usually expressed as its parameter count — “7B” = 7 billion weights, “70B” = 70 billion, “405B” = 405 billion.
Size is the main dial between capability and cost:
| Size | Runs on | Rough feel |
|---|---|---|
| 1–8B | a phone / laptop | fast, lighter |
| 13–34B | a good GPU / Mac | sweet spot locally |
| 70B+ | datacenter-grade | near-frontier |
Bigger usually means smarter, but also slower, hungrier, and pricier to train and run.
Rule of thumb: for local use, ~7–14B is the accessible range; for the best quality you'll go to a cloud API. Size is capability-per-dollar, not absolute quality.