Preprint reports prompting can reduce energy used by AI-generated code

Researchers report in an arXiv preprint that selected prompting strategies reduced the execution energy consumed by AI-generated code by up to 25% for Python and 17% for C++ across the models evaluated. They examined 21 prompting strategies and selected 8 for evaluation across 10 open-weight and proprietary language models against a baseline prompt. Results varied by model and language. For Granite-4.0-H-Small, reductions reached up to 50% for Python and 57% for C++. The study measures the energy used to execute generated code; its reported reductions depend on the prompting strategy, model and programming language.