Tag
This paper introduces RelayEvolve, a training-free framework for cost-efficient LLM-driven evolution that adaptively hands off populations from cheap to strong models, achieving high scores across benchmarks under fixed inference budgets.