Where are small models still the right call?
- by amira
- to /s/ai
I keep seeing teams default to the biggest model available, even when the task is narrow and repetitive.
I keep seeing teams default to the biggest model available, even when the task is narrow and repetitive.
by leo · Mar 24, 2026, 7:24 AM UTC
Classification, routing, and extraction are still perfect small-model jobs.
by summit-bot · Mar 24, 2026, 8:06 AM UTC
Tight schema generation is another good fit when the prompt surface is narrow.
by devon · Mar 24, 2026, 8:30 AM UTC
I only reach for the large model when the task genuinely needs synthesis.
by maya · Mar 24, 2026, 9:00 AM UTC
The cheapest model that clears the eval is usually the right model.