In short
An AI Energy Score rating gives a model one to five stars for energy efficiency compared with other models tested on the same task, based on measured GPU energy per 1,000 queries. More stars means less energy for that task, but the stars are relative within a task, so compare models only against others rated for the same job.
Steps
- Find the task that matches your use, such as text generation or image generation.
- Compare star ratings only between models rated for that same task.
- Look at the measured energy figure as well as the stars.
- Check whether the model was tested with reasoning switched on or off.
- Weigh the rating against quality on your own examples before choosing.
More stars means less energy, but only compared with other models doing the same task. The AI Energy Score, launched by Hugging Face and partners, works a little like an appliance label for AI models.
What does it measure?
Measured GPU energy per 1,000 queries, on a standard set of tasks. The project covers ten common tasks, such as text generation, summarisation and image generation. Each model is run the same way on the same hardware, and given one to five stars relative to other models tested on that task.
How do I use it?
- Pick the right task. A five-star image model and a five-star text classifier are not comparable.
- Read the number, not just the stars. Two models one star apart may be close or far apart in energy.
- Check reasoning. The 2025 refresh added reasoning models and found they used, on average, 30 times more energy; the same model with reasoning on used between 150 and 700 times more than with it off.
- Test quality yourself. A rating tells you about energy, not whether a model is good enough for your work.
What doesn't it tell you?
- Most closed commercial models are not rated, because they cannot be benchmarked from outside.
- It covers GPU energy in a test set-up, not a provider's cooling, idle capacity or water use.
- Your real use differs. Long inputs, long outputs and retries change the energy per task.
Common mistakes
- Comparing stars across tasks.
- Choosing a five-star model that fails your quality bar, then needing a bigger one anyway.
For a broader comparison method, see compare the energy use of different models, and read energy labels for AI.
Sources
- AI Energy Score, Hugging Face (2025)
- Announcing AI Energy Score ratings, Hugging Face (2025)
- AI Energy Score v2: refreshed leaderboard, now with reasoning, Luccioni and Gamazaychikov, Hugging Face (2025)
Related
How do I compare the energy use of different AI models?
How to compare AI models on energy fairly: use benchmarks like the AI Energy Score, or measure them yourself on the same task, hardware and settings.
How do I know if an AI tool is green?
What to look for when judging an AI tool's environmental claims: published energy figures, methods, model size options, hosting and independent ratings.
Energy labels and benchmarks for AI
Where AI energy ratings stand in 2026: AI Energy Score, MLPerf Power, EU AI Act documentation and data-centre rating, and what labels can show.