AI Model CO2e Calculator

AI Model Carbon Emissions Calculator

Select AI Model:

Select Electricity Grid:

About the calculator

This is an experimental calculator exploring the ways to get the latest academic data into the hands of non-experts. The AI Model Carbon Calculator estimates the grams of CO2 for a single mid-length prompt using one of the latest AI models.

To use the calculator, simply select the models from the dropdown and the country most likely to be the location of the data centre where the inference computation is conducted (hint most likely the US at the time of writing but this will change later in 2025).

The main academic source informing the energy use (Wh) per prompt is an excellent paper How Hungry is AI? published in 2025 by the University of Rhode Island which uses a novel benchmarking framework to quantify the carbon footprint of text-based LLM inference models. The detailed analysis uses query and output length, token and latency rates, and the GPU power requirements of different hardware used by the various AI models.

To calculate the grams CO2 equivalent I have multiplied the energy use (in Wh) by the average Scope 2 carbon intensities of the relevant electricity grids (in gCO2/Wh) which are published by the EPA (US) and Defra (UK). As other emissions (Scope 1 direct and Scope 3 supply chain) are not included, the estimate should only be considered a lower bound.

Useful terminology:

Prompt: A short input, such as a question, command, or statement, given to an AI model to guide it in generating a specific response. It provides the necessary context or instructions for the AI to produce relevant and accurate output.
Inference: The process where a trained AI model uses new, unseen data to make predictions, decisions, or draw conclusions. It is the phase when the AI applies what it has learned during training to real-world inputs to generate useful output.