Sustainable AI Group launches tool to estimate proprietary AI emissions

The Sustainable AI Group has released CLEER (Closed-model Latent Energy Estimation Range), a new methodology designed to estimate the energy consumption and greenhouse gas emissions of proprietary AI models. Because major developers such as Anthropic, OpenAI, and Google do not disclose corporate energy data, researchers developed this backdoor method to convert token usage into environmental impact estimates.

Findings from the group indicate significant disparities in energy intensity. Larger, high-capability models, such as Anthropic’s Opus and OpenAI’s Sol, consume nearly four times as much energy on average as smaller models like Haiku or Terra. Furthermore, “agentic” sessions—where an AI performs a series of tasks—use an average of 27 times more energy than standard chat sessions using the same model.

CLEER is being offered as a free, open resource, though companies can license specific datasets for integration into existing carbon accounting systems. E-commerce platform Etsy, along with private equity firm Hg and tax software company Sovos, are among the early adopters of the tool.