AI Energy Score Ratings

This article was written by Sacha Luccioni on Hugging Face.

AI models power everything from chatbots to search engines, but their energy consumption remains largely opaque. Some large models require multiple cloud GPUs to run and use large amounts of energy, whereas other models are small and can run on a laptop or even a phone.

The AI community needs a way to compare the energy consumption of AI models for different tasks – the AI Energy Score project aims to do exactly this, offering a clear and standardized benchmark for measuring AI energy consumption, ensuring that the AI community can make informed, sustainable choices.

Key Features of the AI Energy Score

Standardized Energy Ratings ⚡

A uniform framework measures and compares AI models based on their energy consumption during specific tasks. This ensures clarity and reliability in evaluating model efficiency.

Public Leaderboard of AI Models 📈

For the first time, a publicly available leaderboard ranks 166 AI models across 10 common AI tasks—including text and image generation, summarization, and more. Among the models evaluated are the LLaMa models, Phi, Gemma, Mistral and SmolLM. 

A Benchmarking Portal for AI Developers 💻

Developers can submit their models—whether open-source or proprietary—to be assessed under the AI Energy Score framework. Open models are automatically tested, while closed models undergo evaluation via a secure Docker environment, ensuring reproducibility and a fair comparison.

Recognizable Energy Use Labels ⭐

To simplify decision-making, the AI Energy Score introduces a 1- to 5-star rating system, similar to how household appliances are rated for energy efficiency. The most efficient models for a given task receive the top “five star” rating, guiding users toward sustainable choices.

Please click on this link to read the full article.

Image credit: Image by Freepik