SpecEvo paper proposes multi-LLM framework to cut scientific discovery costs by up to 80%

Researchers propose Speculative Evolution (SpecEvo), a multi-LLM framework that coordinates lightweight and frontier models to reduce the cost of LLM-driven evolutionary search for scientific discovery. The approach delegates routine exploration and evidence consolidation to inexpensive models while reserving frontier reasoning for strategic interventions.

The framework uses lightweight models as parallel Speculators to generate and refine candidate programs and as an Advisor to distill accumulated successes, failures, and saturated directions into persistent guidance for subsequent exploration.

At sparse checkpoints, a frontier-model Navigator inspects the evolving population and intervenes through synthesis, surgical refinement, or reframing when search stagnation is detected, enabling targeted use of expensive reasoning.

Across 176 tasks spanning four scientific and algorithmic discovery domains, SpecEvo matches or outperforms frontier-model baselines while reducing API cost by up to 80%.

On equation discovery, the method achieves up to 1.5x greater evaluation efficiency than cost-efficient baselines, and it improves over the strongest CO-Bench baseline by 3.8%.

The authors argue that trajectory-conditioned model coordination is an effective approach to cost-efficient scientific discovery, supporting scalable AI-driven research workflows.