FrugalEvo: Towards Cost-Aware LLM-Guided Program Evolution
arXiv cs.AI · 3 Oct 2026, 03:44 AEST
FrugalEvo introduces a cost-aware evolutionary framework for optimizing computational problems. It outperforms state-of-the-art methods in solution quality while being more cost-effective.
Key facts
- LLM-guided evolutionary methods, such as AlphaEvolve, have emerged as powerful approaches for challenging computational optimization problems, such as circle packing.
- We propose FrugalEvo, a cost-aware evolutionary framework where a stronger, higher-cost LLM explores solution strategies, and a cheaper LLM implements them and iteratively refines the resulting code.
- Across 10 mathematical and systems optimization tasks, FrugalEvo matches or surpasses state-of-the-art baselines, including OpenEvolve, ShinkaEvolve, AdaEvolve, and EvoX, in final solution quality and achieves higher BA-AUC on 9 tasks.
- Notably, on circle packing, FrugalEvo achieves new state-of-the-art performance with GPT-5.6 Terra and Luna for only 1.68 USD and with GLM-5.3 and its Flash variant for only 0.55 USD.
Read aloud
FrugalEvo introduces a cost-aware evolutionary framework for optimizing computational problems. It outperforms state-of-the-art methods in solution quality while being more cost-effective. L L M-guided evolutionary methods, such as AlphaEvolve, have emerged as powerful approaches for challenging computational optimization problems, such as circle packing. We propose FrugalEvo, a cost-aware evolutionary framework where a stronger, higher-cost L L M explores solution strategies, and a cheaper L L M implements them and iteratively refines the resulting code. Across ten mathematical and systems optimization tasks, FrugalEvo matches or surpasses state-of-the-art baselines, including OpenEvolve, ShinkaEvolve, AdaEvolve, and EvoX, in final solution quality and achieves higher BA-AUC on nine tasks. Notably, on circle packing, FrugalEvo achieves new state-of-the-art performance with G P T five.six Terra and Luna for only one point six eight USD and with GLM-five point three and its Flash variant for only zero point five five USD.
Sources in this cluster
- FrugalEvo: Towards Cost-Aware LLM-Guided Program Evolution
arXiv cs.AI · 3 Oct 2026, 03:44 AEST