Template · Multi-model reasoning
Smart Escalation
A fast model answers first. If the answer is uncertain or incomplete, stronger models are consulted and their answers are synthesized.
When to use it
Best when you want to control cost without giving up quality. The cheap path handles easy prompts, the expensive path handles hard ones.
The research behind it
This template is based on FrugalGPT (Chen et al., 2023).
Related templates
Consensus Draft
Three models answer your question independently, then a synthesizer combines the strongest points into one response.
FreeRank & Fuse
Three models generate candidate answers, a ranker scores them, then a fuser combines the strongest elements guided by the ranking.
FreeBest-of-N
Five high-temperature responses are generated, scored, and the best one is selected.
Run Smart Escalation
Add your prompt and let the template do the rest. Free to start, no card.
Open in LLMWeave