Local models are useful when wrong answers are cheap to catch

A model-choice note about using local models for small verifiable tasks and stronger models for high-cost reasoning.

Local models can be very useful for narrow coding tasks: boilerplate, regex drafts, log summaries, small test ideas, or explaining a function. I trust them more when the answer is cheap to verify. For long-range reasoning, current API behavior, security-sensitive review, or multi-file architecture calls, the cost of a wrong answer is higher and a stronger model may be worth it.