Summarize this article with:
GreenPT joins Eden AI as a green alternative: energy and carbon measured on every request, compression models that cut output tokens, and private, open-weight models behind one API.
Most AI providers have a sustainability page. Very few put a number in the response.
That gap is the whole problem with "green AI" today. Training a model is a one-off event; inference runs every time a user sends a request, for as long as the product exists. As AI features move into production, their footprint stops being a research question and becomes an operating cost, one that almost nobody can see.
GreenPT is built around making it visible, and it is now available on Eden AI.
Who GreenPT is
GreenPT is a Dutch AI company with two commitments: your data is never used to train a model, and the environmental cost of AI is measured rather than claimed. Its API serves open-weight models, from its own GreenL and GreenR to GLM, DeepSeek, Kimi and MiniMax, through an OpenAI-compatible interface. GreenPT tells its side of the partnership in its own announcement: GreenPT is now available on Eden AI.
Lever 1: measure every request
According to GreenPT, the energy of each request is read directly from the GPU hardware that served it, and the carbon figure uses the live carbon intensity of the grid that GPU is connected to. No annual averages, no renewable certificates standing in for the real number.
That is what makes the rest possible. A team that has to report on the footprint of its AI features can log energy and carbon per request, next to latency and cost, and see which feature, customer or model actually drives it.
Through Eden AI, these figures come back in every GreenPT response as an impact object, with inference time, energy in watt-milliseconds and estimated emissions in micrograms of CO2e. No other provider on Eden AI returns this today.
It also explains why the numbers move. Two requests on two different grids, or at two different times of day, will not have the same carbon figure. That is not noise to hide; it is the honest version of the measurement.
Lever 2: generate fewer tokens
The cleanest token is the one that is never generated. GreenPT's compression models are variants of GLM 5.2 that carry a built-in instruction to answer more briefly, at the same price per token:
cavemantrims proseponytailtrims generated codehoneydoes both
Each comes in a -lite and an -ultra intensity. On its own test set, GreenPT measured output reductions of roughly a third to three quarters against the base model, and most of the saving came from shorter reasoning and explanation, not shorter code. Fewer output tokens means less compute per answer, and a smaller bill at the same time.
There is a catch, and GreenPT documents it: the instruction itself costs a few hundred input tokens. On short answers such as classification, routing or extraction, a compression model costs more than the base model. It pays off on long answers: coding agents, documentation, support replies.
Lever 3: the right model, and nothing kept
Not every question needs a heavyweight model. GreenPT's own GreenL and GreenR carry a built-in sustainability system prompt, while their -raw twins carry none. For new projects, GreenPT points to GLM 5.3 and GLM 5.3 Flash as its recommended chat models, and its own Green S models handle speech-to-text.
On data, GreenPT states that it never trains on your inputs and keeps nothing beyond short-lived operational caching. Through Eden AI, request content is not stored by default either, and GreenPT's models are tagged EU and zero data retention in the catalogue.
Try it through Eden AI
On Eden AI, GreenPT sits behind the same OpenAI-compatible endpoint as every other provider. The full list of GreenPT models is on the GreenPT provider page. Moving your longest-answer workload to a compression model is a one-line change:
curl https://api.edenai.run/v3/chat/completions \
-H "Authorization: Bearer $EDENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "greenpt/glm-5.2-honey",
"messages": [{"role": "user", "content": "Add retry with backoff to this fetch call."}]
}'Swap greenpt/glm-5.2-honey for greenpt/glm-5.3, greenpt/green-l or any other GreenPT model, and compare output length, cost and quality on your own prompts before moving anything to production. One key and one bill cover GreenPT and every other provider on Eden AI.
What we think
"I think the footprint of AI is the question our industry keeps putting off. What I like about GreenPT is that they don't just promise green: their infrastructure runs on renewable energy in Europe, and every request tells you what it actually cost. That's the kind of transparency we want to offer our users."
Samy Melaine, Co-founder, Eden AI
"Now that GreenPT is on Eden AI, switching to AI that runs on renewable energy and keeps your data in Europe doesn't mean rebuilding anything. You keep the API you already use and change one line. Eden AI cares about transparency as much as we do, which made this an easy partnership to build."
Robert Keus, CEO, GreenPT
Where to start
Pick one workload with long answers, a coding agent or a support assistant, and run it on a GreenPT compression model for a week. Compare the output tokens and the bill with what you run today. If the answers hold, you have cut energy and cost at the same time, without changing anything else in your stack.
