FrugalEvo: Towards Cost-Aware LLM-Guided Program Evolution
FrugalEvo pairs a costly strategy model with cheap code refinement and beats evolutionary baselines at far lower spend.
FrugalEvo is a cost-aware evolutionary framework in which a stronger LLM explores solution strategies and a cheaper LLM implements and iteratively refines code, with prompts designed to reuse cached prefixes. Across 10 mathematical and systems tasks it matches or surpasses OpenEvolve, ShinkaEvolve, AdaEvolve, and EvoX in final quality and records higher BA-AUC on 9 tasks. On circle packing it reports new state-of-the-art results for 1.68 USD with GPT-5.6 Terra and Luna, or 0.55 USD with GLM-5.3 and Flash, versus about 50 USD for multi-agent baselines such as CORAL and SwarmResearch.
- A costly LLM searches strategies while a cheaper model writes and refines code.
- Budget-Aware AUC scores quality against cumulative LLM cost.
- Circle packing reaches reported SOTA for 1.68 USD or 0.55 USD.
- Higher BA-AUC on 9 of 10 math and systems tasks.
- AI models
- GPT-5.6 TerraGPT-5.6 LunaGLM-5.3
From the original article
1 min read at huggingface.co
LLM-guided evolutionary methods, such as AlphaEvolve, have emerged as powerful approaches for challenging computational optimization problems, such as circle packing. However, prior work typically optimizes performance gain over a fixed number of iterations. We argue that practical optimization should maximize gain per unit cost.
By Hui Chen, Xuan Qi, James Xu Zhao · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.