Confier une tâche à une IA spécialisée

Name: Give a task to a specialised AI
Creator: Projet Celsius
License: https://creativecommons.org/licenses/by/4.0/

Projet Celsius

Database entryUsesLevel 3 of 6 useUse, level 3 of 6Used to calculate 3 entries, levels 4 and 5Calculated with 1 entry, level 1Database structure

Give a task to a specialised AI

A task given to a specialised model (classify, summarise, detect), smaller than a conversational assistant, measured on open models.

Selected reference: Classify a text, source Luccioni et al.
CO2e is calculated from electricity and the worldwide grid; it is not a direct measurement of carbon.
Out of scope: cooling of the data centre, server manufacturing and model training.

Compare

Calculated with

Grid electricity

Used to calculate

Annual emissions for giving a task to a specialised AI, by frequency

Your frequencyPeriod

Read the values at100 times a year

Reading: at 490,000 uses per year, "Specialised task" weighs as much as one phone charge every week. Carbon footprint (CO2e) over one year, by number of uses. Each use is a straight line; "Specialised task" is highlighted, with the chosen parameters, compared with 6 reference uses at their default settings; the solid line shows one phone charge per week over a year, and the point where the task reaches it. Servers worldwide. Calculation from the AI Impact Database by Projet Celsius.

Origin of the value

Classifying a text costs about 1,500 times less than generating an image in the same study.

Comparison with the other uses

The topic "Code and agents"

Specialised task0.00092 gCO2e
Code suggestion0.031 gCO2e, 34 times the value of "Specialised task"

↓ The heaviest use in the topic, "Coding agent hour": 310 gCO2e, 340,000 times more

Emissions of one use, linear scale, 2 of the 7 uses in the topic. For calculated values, servers worldwide.

Characteristics

Unit	1 task : classify a text
Default location	world, 458 gCO2e/kWh (Ember 2025, life cycle); 11 other locations to choose from
Counted	the server electricity measured over 1,000 inferences, divided by 1,000
Not counted	cooling of the data centre, server manufacturing and model training
Status	measured, 1 recorded value of the same scope, 4 other published values

All the characteristics identifier, version, rule for the value, sources, licence

Identifier	IA-0050, projetcelsius.com/en/ai/database/IA-0050/
Version	0.3 of 2 November 2026
Category	Uses, a request to AI
Task	Classify a text (0.00092 gCO2e), Classify an image (0.0032 gCO2e), Detect objects in an image (0.017 gCO2e), Generate a short text (0.022 gCO2e), Summarise a text (0.022 gCO2e)
Choice of the selected value	value taken as it stands from its source, without calculation
Sources	Luccioni et al., 2024
Licence	CC BY 4.0

Emissions by server location

Emissions associated with electricity (CO2e)

Poland0.0012 gCO2e
World0.00092 gCO2e
United States0.00077 gCO2e
European Union0.00042 gCO2e
France0.000083 gCO2e

× 14 between Poland and France, for the same electricity

Ember 2025, over the life cycle. The same use consumes 0.002 Wh in each location.

Giving a task to a specialised AI · result per task · measured · one server location per line.Counted: the server electricity measured over 1,000 inferences, divided by 1,000. Out of scope: cooling of the data centre, server manufacturing and model trainingFind this case: projetcelsius.com/en/ai/database/tasks/task-for-a-specialised-ai/Projet Celsius calculation, AI Impact Database, version 0.4 of 13 November 2026 · IA-0050, entry added in version 0.3 · engine 0.3.1 · CC BY 4.0.

Result for the chosen case

Location of the servers

Task

For a carbon footprint

The row for a carbon footprint

Item	purchases of services (GHG Protocol scope 3, category 1); regulatory BEGES, category 4
Quantity	1 task
Factor	0.00092 gCO2e per task, IA-0050 version 0.3, based on the local grid (location-based), servers worldwide
Total	0.00092 gCO2e · not counted: cooling of the data centre, server manufacturing and model training

This block is made of no other: its value is read from its sources.

Entries that use it (3)

Automated monitoring, 1 year, Process automation, Data scientist, 1 year

Sources of the calculation

2 publications, for this entry and those used to calculate it, with the chosen parameters.

Energy and factors

The references of the selected values and of the entries used in this calculation.

Luccioni et al. · October 2024
Power Hungry Processing ↗ (new tab)
Specialised task
Role and limits : Luccioni et al.
Author. Luccioni, Jernite and Strubell
What the publication covers. GPU, processor, memory
What it does not cover. data centre
Reading by the database. Open models only
- Specialised task →
  Reference for the "Classify a text" variant
  Scope of the figure used. Text classification, average of the open models tested: 0.002 kWh for 1,000 inferences
  Limit. Models of 2023, measured on a node of 8 A100 GPUs

Conversion into carbon

The source of the electricity factor of the selected location.

Worldwide: 458 gCO2e/kWh, life cycle.

Ember · 2026
Global Electricity Review 2026, and Our World in Data based on Ember ↗ (new tab)
Grid electricity
Role and limits : Ember
Author. Ember, Our World in Data
What the publication covers. electricity generation over the life cycle, by country
What it does not cover. guarantees of origin
Reading by the database. The same source for all countries
- Grid electricity →
  Conversion of electricity into carbon: worldwide
  Scope of the figure used. the production of electricity over the life cycle, including power plant construction and fuels
  Limit. An average for the chosen country, not the electricity contract or the data centre of a specific service.

Published values

5 figures for the same operation, each within its own scope.

0.002 Wh selected, between 0.002 Wh and 0.049 Wh for published values of the same scope

Green: same scope as the selected value (measured in dark, estimated or published in light). Grey: other scope. Empty circle: check by the database or discarded figure. Logarithmic scale.

Source	Date	Electricity	What the figure measures	Scope	Status
Specialised taskselected by the database, v0.3	2026	0.002 Wh	the value selected by the database	same scope	measured
Luccioni, Jernite and Strubell ↗Power Hungry Processing	Oct. 2024	0.002 Wh	Text classification, average of the open models tested: 0.002 kWh for 1,000 inferences · Models of 2023, measured on a node of 8 A100 GPUs	same scope	measured
Luccioni, Jernite and Strubell ↗Power Hungry Processing	Oct. 2024	0.007 Wh	Image classification, average of the open models tested: 0.007 kWh for 1,000 inferences · Models of 2023, measured on a node of 8 A100 GPUs	same scope	measured
Luccioni, Jernite and Strubell ↗Power Hungry Processing	Oct. 2024	0.038 Wh	Object detection, average of the open models tested: 0.038 kWh for 1,000 inferences · Models of 2023, measured on a node of 8 A100 GPUs	same scope	measured
Luccioni, Jernite and Strubell ↗Power Hungry Processing	Oct. 2024	0.047 Wh	Text generation by specialised models, 1,000 inferences on a GPU: 0.047 kWh · Not conversational assistants: smaller models, on short tasks	same scope	measured
Luccioni, Jernite and Strubell ↗Power Hungry Processing	Oct. 2024	0.049 Wh	Text summarisation, average of the open models tested: 0.049 kWh for 1,000 inferences · Models of 2023, measured on a node of 8 A100 GPUs	same scope	measured

The published figures do not all count the same scope; the scope of each figure is written next to it. Why do the estimates diverge?

Methodology conventions

With their alternatives

The full methodology

The default location	Without other information, the servers are in the world, at the world average electricity · alternatives: the United States, where most of the large assistants run
Electricity, location-based, on a life cycle basis	The kilowatt-hour is that of the servers' country, on a life cycle basis, from a single source for all countries · alternatives: market-based, with purchase contracts, shown alongside when published
Two significant figures	A calculated value is rounded to two significant figures, a published value is reproduced with the precision of its source; the calculation keeps full precision
Calculate a case without an entry	The formula, the default values and a step-by-step example

Uses in the hub

Charts, figures, analyses and articles that cite this entry

Analysis"A generated image is a phone charge" →
ChartElectricity by type of AI task →

History

Entry in version 0.3 of 2 November 2026, in the database in version 0.4

The changelog

2 November 2026Designation : A specialised AI task becomes Specialised AI task (classification).More precise designation: the object first, the specification after the comma, the quantity or period at the end; the former title remains searchable.
2 November 2026Designation : Specialised AI task (classification) becomes Giving a task to a specialised AI.The name in people's own words, without a leading article.
2 November 2026Address : tache-d-ia-specialisee becomes confier-une-tache-a-une-ia-specialisee.In people's own words; the old address redirected.
1st November 2026Creation : version 0.1.Entry added: a specialised AI task, five tasks measured by Luccioni, Jernite and Strubell.
31 October 2026Creation of the database, in its first version (0.1).

Citation

Projet Celsius, AI Impact Database, version 0.4 of 13 November 2026, "Giving a task to a specialised AI", IA-0050 (entry added in version 0.3 of 2 November 2026), projetcelsius.com/en/ai/database/IA-0050/ · CC BY 4.0.

Download

JSON Markdown