Standardise environmental impact metrics for efficient and green AI models
Beyond State-of-the-Art: Standardising Environmental Impact Metrics for AI Research
Artificial Intelligence
Summary
Training and running large AI models use a lot of electricity, which can harm the environment. The authors found almost no papers report how much energy or carbon emissions their AI models cause. They created standard ways to measure and report this impact, along with a software tool to help researchers do it easily. They also introduced a new idea to encourage using the smallest AI model that meets the goal, instead of always chasing the very best accuracy with huge energy costs.
What this means in practice
- •For machine learning teams: Track and report carbon emissions from AI model training using a plug-and-play software tool based on standardised sustainability metrics.
- •For cloud platform operators: Encourage efficient AI models by adopting carbon impact metrics that incentivise using smaller models for tasks to reduce data center emissions.
Authors
Lachlan McGinness, Dan Pagendam, Robert Offner
Abstract
As the capabilities and ubiquity of Large Language Models (LLMs) grow, so does their environmental footprint. Despite calls for responsible AI, the machine learning community lacks standardised practices for carbon accounting. Our automated literature review of the 5,285 papers accepted to NeurIPS 2025 reveals that reporting of environmental impact is nearly non-existent. To catalyse a shift toward sustainable AI, we define standardised sustainability metrics for evaluating model training efficiency, accompanied by simple heuristics to estimate the carbon cost of LLM inference. We implement these metrics in carbonbenchmark, a drop-in software solution for tracking and reporting emissions. Finally, to combat the pursuit of marginal accuracy gains at disproportionate environmental costs, we formalise the `Smallest Model that Achieves the Job' (SMAJ), a framework which challenges the field to prioritise computational efficiency and environmental accountability alongside traditional `State-of-the-Art' (SotA) accuracy.