Term: Parameters
~5 min read
Estimated time: ~5 min read — for the in-app brief plus opening the primary source.
What this is
Parameters are the learned weights inside a model. People use size (for example “70B”) as a shorthand for how large — and often how costly — the engine is.
Everyday example
A vendor says 'our 70-billion-parameter model.' That number is the size of the learned engine — a rough signal of capacity and cost, not a quality guarantee for your contracts or tickets.
Parameters are the learned knobs inside a model — a size signal, not a quality guarantee.
- Bigger is not automatically better for your task.
- Vendors quote parameter counts as prestige; your evals matter more.
- Cost and latency often rise with size.
- Ask what the model is good at, not only how large it is.
Next action: Ignore parameter bragging until you see results on your own sample of work.
What changes in how you lead
How decision rights, process, and ownership should change.
- Size is not a procurement criterion on its own.
- Finance should see unit cost on your tasks, not a lab’s headline number.
Compare related ideas
Parameters vs Frontier model
Parameter count is size. Frontier is a capability tier at a point in time. A smaller model can outperform a larger one on a narrow, well-defined task.
Open Frontier modelDeep dive
Think of parameters as the knobs the model learned during training. More knobs can capture more nuance — and cost more to run.
Frontier chat products rarely lead with a parameter count; smaller or open models often do. Either way, eval on your work.
High-volume classification or short rewrites often run well on a smaller model. Hard reasoning and long drafting may need a frontier tier.
Purchasing question: which model size for which process, and can we mix tiers?