Generating Energy-Efficient Code via Large-Language Models - Where are we now?

24-Sep-2025

We are excited to announce that a paper by the Software and Sustainability research group and colleagues has been accepted at the International Conference on Software Engineering (ICSE 2026), the premier conference in software engineering!

Title: Generating Energy-Efficient Code via Large-Language Models – Where are we now?

Authors: Radu Apsan, Vincenzo Stoico, Michel Albonico, Rudra Dhar, Karthik Vaidhyanathan, and Ivano Malavolta.

🔗 Read the paper here

💡 Key Findings

• 🟢 Human expert code remains the most energy-efficient, outperforming all LLM-generated code by 17% to 30%.
• 🤖 LLMs outperform average human developers on some platforms, but still lag behind green-expert solutions.
• ⚙️ Prompt engineering does not consistently improve energy efficiency.
• 🌩️ Human expertise is still crucial for developing energy-efficient Python code.