How should prompts differ for small versus large models?
Small models benefit from tighter constraints, explicit steps, and more examples. Large models can infer more but may wander—keep formats strict. For small models, avoid ambiguity and reduce cognitive load: shorter context, clear schemas, and simple tasks. For large models, use staged prompts and self-checks to channel capability.
In practice
If a small model fails, try adding one crystal-clear example before switching models.
Where to start
- [01]Create ‘small model’ and ‘large model’ prompt variants.
- [02]Measure accuracy vs. cost across both.
- [03]Route tasks by complexity to the right tier.
- [04]Re-evaluate routing quarterly.
small model promptingLLM size differencesprompting tiny models