Coroid provides an estimate of the carbon emissions caused by the model inference used in your work, and offers a Low carbon AI profile that steers tasks towards lower-impact options.
What the estimate covers
Operational inference only — the energy used to run models for your work, along with the carbon intensity of the grid serving the region where the model operates.
What it does not cover
Defining the exact scope is more important than the numerical value:
- Not model training. The substantial one-off cost of creating a model is excluded.
- Not embodied carbon. Manufacturing the hardware is not included.
- Not your own infrastructure. Your CI processes, test runs, and deployments.
- Not a claim of carbon neutrality. Nothing here indicates that emissions are offset.
It is an estimate of a specific, real quantity. Use it as a comparative indicator — helpful for spotting which configuration has a higher impact — rather than as an audited, absolute figure.
The Low carbon profile
This profile favours smaller models running on grids with lower carbon intensity. Both factors matter: a smaller model performs less work per request, and the same task generates different emissions depending on the location.
There is a genuine trade-off here. Smaller models have reduced capabilities across the entire pipeline, so the Low carbon profile works better for routine tasks than for architecturally critical work. Select it based on each task rather than using it as a default setting — see AI profiles.
Cutting impact without altering profiles
The biggest reduction available to most teams is not related to model choice — it lies in avoiding unnecessary work:
- Reduce rework. A task that runs twice naturally produces double the emissions. Clear specifications are the key lever. See Specifications.
- Define task scopes precisely. Broad tasks explore more and thus set their own boundaries.
- Keep context minimal. Context makes up most of the tokens used.
These steps also cut costs, as both expenses and emissions depend on the number of tokens.
Where to view this information
Model profiles display carbon characteristics alongside speed, quality, and cost, enabling you to consider all four factors when choosing a profile.