Sustainable AI Group has released CLEER — Closed-model Latent Energy Estimation Range — a free, open methodology that converts a company's token usage, the text processed by enterprise AI models from vendors such as Anthropic, Google, Microsoft and OpenAI, into energy and greenhouse gas emissions estimates.

The company behind the tool was co-founded by former Salesforce sustainability manager Boris Gamazaychikov, who serves as CEO. He says the resource targets the full range of models a corporate sustainability team typically licenses.

"In the case of a sustainability professional, pretty much every one they use would apply," Gamazaychikov said.

E-commerce marketplace Etsy helped design and test CLEER and is one of its first customers, alongside private equity firm Hg and tax software company Sovos. Etsy plans to license specific CLEER datasets for ongoing use in its existing carbon accounting systems, even though the core methodology is free.

"When new environmental challenges emerge, we don't wait for the industry to give us new answers," said Chelsea Mozen, head of impact and sustainability at Etsy. "We test new approaches, learn what works and share those insights so others can move forward with us."

Built on proxy data from more than 20 open models

CLEER addresses a gap in corporate carbon accounting: emissions from proprietary AI models. The model providers do not disclose energy or emissions data for the systems companies license, so Sustainable AI Group built proxies using disclosed information for more than 20 open models. It published this earlier work in the AI Energy Score resource.

The methodology is designed to improve as disclosure expands. When a model provider publishes trustworthy model-specific measurements, those actual figures can replace the corresponding CLEER estimates without requiring users to rebuild their accounting approach, according to Sustainable AI Group.

Etsy brings prior form to this problem. The company previously built and published its own method for estimating emissions from its cloud computing services, Cloud Jewels, so other companies could apply it. Etsy is also one of the few companies that explicitly breaks out emissions for its digital operations within its broader disclosures for purchased goods and services.

Model choice can move energy consumption by more than 30 percent

CLEER functions as more than a measurement methodology. It also allows companies to compare the relative energy performance of competing AI models — and the spread is material. Sustainable AI Group found that model choices for certain tasks can affect energy consumption by more than 30 percent, according to the CLEER technical analysis.

The analysis also found a pattern with direct cost implications: more often than not, the cheaper AI models carried higher estimated energy needs. For procurement teams weighing AI vendor contracts, that inverts the usual price signal — the lowest per-token cost may carry the highest emissions burden per unit of work.

That finding gives sustainability managers a defensible basis for介入 model selection conversations that have traditionally sat with engineering and procurement. It also gives companies that report under Scope 3, purchased goods and services, a way to attribute AI-related emissions now rather than waiting for vendor disclosure regimes to mature.

Watershed integration in the works for some 800 companies

Distribution is the next hurdle, and Sustainable AI Group is pursuing it through the carbon accounting software stack. The company is building a CLEER integration for Watershed, a platform used by about 800 companies. Watershed is separately developing a complementary framework to inform how companies account for AI emissions, and the two companies will collaborate.

Sustainable AI Group is exploring additional software integration partnerships beyond Watershed.

"We want this data to go into the places where sustainability professionals are already doing their accounting," Gamazaychikov said.

Estimates, not disclosures — for now

The methodology's accuracy depends on proxy data, and Sustainable AI Group itself frames accuracy as a function of adoption: the resource becomes more accurate as more organizations use it and as model providers disclose actual figures. Until then, CLEER outputs are estimates, clearly labeled as a range rather than point measurements.

For the companies already using it — Etsy, Hg, Sovos — the value proposition is immediacy. AI workloads are scaling fast inside enterprise operations, and purchased electricity and services categories in corporate inventories are growing accordingly. CLEER gives sustainability teams a first-pass number for a stream that has so far been a black box in their disclosures.

What happens next hinges on two developments: whether Watershed ships the integration and complementary accounting framework to its roughly 800 customers, and whether major model vendors begin publishing model-specific energy data that can replace CLEER's proxies. Either would move AI emissions from estimation toward measurement — and both are milestones worth tracking through the next reporting cycle.