Carbon reduction
How much CO₂e Zyggs can avoid in a year, with the arithmetic open, the source of each number and its uncertainty. Change the scenario and watch the result change.
In short
- Two terms: assistant requests answered on the computer, and drawings that would have gone on paper.
- Every physical number comes with its source and range; the usage numbers are the scenario you choose.
- An answer on the computer spends less than a tenth of the energy estimated for the cloud (measured).
- On a clean grid such as Brazil's, paper weighs most; on a coal-heavy grid, the gain from paper can vanish.
Your scenario
An estimate, not a measurement: the usage numbers are the scenario you choose.
80% of the simulations between 25.2 and 51.3 t
3.66 kg CO₂e per person
The total is positive in 100% of the simulations.
20,000 draws, seed 20261107: the same numbers on every visit.
What weighs most
The yearly tonnes when each number goes from its 10th to its 90th percentile, with the others at their most likely value.
| Number | At its 10th percentile | At its 90th percentile |
|---|---|---|
| Carbon of one A4 sheet | 24.2 t | 47.8 t |
| Energy of one answer in the cloud | 28.1 t | 38.1 t |
| Power of the computer and screen | 28.6 t | 26.2 t |
| Carbon of the computer's grid | 28.1 t | 25.8 t |
| Minutes per drawing | 28.7 t | 26.6 t |
The question
How much CO₂e stays out of the atmosphere in a year when people use Zyggs instead of what they would use without it. It is an estimate made with open arithmetic, not a measurement: the program collects no usage statistics, so how much it is used is the scenario you choose in the calculator.
Term 1: the assistant answers on the computer
Every request to Zyggs's assistant is answered on your computer. Without Zyggs, part of those requests would go to a language model in a data centre. The arithmetic compares the two energies, each with the carbon intensity of the grid that supplies it:
- : days of use a year; : assistant requests per day of use; : share of those requests that would go to the cloud without Zyggs.
- : energy of one answer in the cloud (Wh); : carbon of the data centre’s grid (g CO₂e/kWh).
- : energy of the answer on the computer (Wh), the measured time multiplied by the processor’s extra power; : carbon of the grid where the computer is.
The time was measured: the assistant of Zyggs 3.14.0 answered 15 requests on a Core i7-13650HX in 3.4 s at the median (4.0 s at the 90th percentile), one process per request, as the program does. With 25 W more on the processor, one answer spends about 0.024 Wh, less than a tenth of the 0.3 Wh estimated for the cloud.
Term 2: drawings that would have gone on paper
In traditional animation each drawing takes a sheet. For the share of drawings that, without Zyggs, would be made on paper, the arithmetic adds the carbon of the sheet and takes away the energy of the computer and its screen during the drawing's time:
- : drawings per day of use; : share of them that would go on paper.
- : carbon of one A4 sheet of 80 g/m² (g CO₂e).
- : minutes on the computer per drawing; : power of the computer and screen (W).
This term can turn negative: on a coal grid, a computer on for 15 minutes emits about as much as a sheet. Nothing in the arithmetic stops the result from being zero or negative.
The numbers and where they come from
| Number | Minimum | Most likely | Maximum | Source |
|---|---|---|---|---|
| Energy of one answer in the cloud | 0.1 Wh | 0.3 Wh | 2.5 Wh | Epoch AI, 2025: 0.3 Wh for a typical GPT-4o query; long inputs reach 2.5 Wh. |
| Carbon of the data centre’s grid | 445 g CO₂e/kWh | 459 g CO₂e/kWh | 473 g CO₂e/kWh | World average of 2024: 445 (IEA, Electricity 2025) to 473 (Ember, 2025). |
| Time of one answer on the computer | 2.5 s | 3.4 s | 6 s | Measured on 07/11/2026: Zyggs 3.14.0 on a Core i7-13650HX, 15 requests, median 3.4 s, 90th percentile 4.0 s, slowest 6.0 s. |
| Extra power of the processor | 15 W | 25 W | 45 W | Assumption: one or two busy cores of a laptop processor. |
| Carbon of one A4 sheet | 3 g CO₂e | 4.7 g CO₂e | 9.9 g CO₂e | Tomberlin, Venditti and Yao, 2020 (252 US mills): 942 kg CO₂e per tonne on average, 608 to 1978 by grade; a sheet of 80 g/m² weighs 5 g. |
| Minutes per drawing | 5 min | 15 min | 30 min | Assumption: from a quick in-between to a careful key drawing. |
| Power of the computer and screen | 15 W | 40 W | 90 W | Assumption: a laptop at the low end, a desktop with a monitor at the high end. |
| Carbon of the computer's grid (Brazil) | 28.9 g CO₂e/kWh | 38.5 g CO₂e/kWh | 92 g CO₂e/kWh | Brazil: 38.5 in 2023 (MCTI, annual), 28.9 in April 2025 (MCTI, monthly), 92 in 2021, a dry year (IEMA, another method). World: as for the data centre. |
| Carbon of the computer's grid (World average) | 445 g CO₂e/kWh | 459 g CO₂e/kWh | 473 g CO₂e/kWh | Brazil: 38.5 in 2023 (MCTI, annual), 28.9 in April 2025 (MCTI, monthly), 92 in 2021, a dry year (IEMA, another method). World: as for the data centre. |
How uncertainty enters
Each uncertain number is a triangular distribution: minimum, most likely value and maximum, as in the table. The calculator draws 20,000 combinations with a fixed seed (always the same draws) and shows the 10th percentile, the median and the 90th percentile of the result. The sensitivity table moves one number at a time between its 10th and 90th percentile, with the others at their most likely value, and shows what weighs most.
The arithmetic has automatic tests: units checked by hand, the triangular sampling, the seed, the order of the percentiles, proportionality to the number of people and the absence of any cut at zero.
What is left out
- The training of the assistant's model, done once, and the cloud space for projects.
- Downloads of the installer and the hosting of the site.
- The manufacturing of the computers.
- Motion capture: compared with a cloud service it can spend more or less energy, and we have no reliable measure of it.
- On the paper side, the light table, the scanner, the pencil and the disposal of the sheet. Leaving them out lowers the calculated benefit.
What this number is not
- Not a measurement: the program collects no usage statistics.
- Not a carbon credit or an offset.
- It does not count rebound effects, such as drawing more because it is easier.
Sources
- Epoch AI, 2025 How much energy does ChatGPT use?
- IEA, 2025 Electricity 2025: Emissions (445 g CO₂/kWh in 2024)
- Ember, 2025 Global Electricity Review 2025 (473 g CO₂e/kWh in 2024)
- MCTI, SIRENE Emissões e fatores de emissão do Sistema Interligado Nacional
- eixos, 2024 Brasil tem menor taxa de emissão de carbono na geração elétrica em onze anos (MCTI: 38,5 kg CO₂/MWh em 2023)
- Canal Solar, 2025 Emissões de CO₂ da geração elétrica (MCTI: 28,9 kg CO₂/MWh em abril de 2025)
- eixos, 2023 Termelétricas geram menos e emissões caem 65% em 2022, mostra estudo (IEMA: 92 e 32 t CO₂e/GWh no SIN em 2021 e 2022)
- Tomberlin, Venditti, Yao, 2020 Life cycle carbon footprint analysis of pulp and paper grades in the United States using production-line-based data and integration (BioResources 15(2))