Performance of AI models
On 10 November 2026, among the 7 models whose electricity use can be estimated, Claude Opus 5.5 completes the most coding tasks (57.8%). It also needs the most electricity per task: 200 to 610 Wh, as much as about 670 to 2,000 quick questions; Claude Haiku 5.5 needs the least: 13 to 38 Wh.
Coding tasks completed and electricity use of a task, 7 models
CursorBench benchmark, best setting of each model; electricity use of a task in Wh, estimate, logarithmic scale; data as of 10/11/2026
Bar: the electricity use of a task, from "all tokens read" to "all written" (estimate)In deep green, the model that completes the most tasks
Sources: Epoch AI, CursorBench; electricity: Projet Celsius estimate, AI Impact Database (IA-0011). Licence CC BY 4.0.

Estimation methodology
CursorBench benchmark, best setting of each model; only models whose size is known (written in the name) or deduced from a price checked with the company are estimated. Electricity: Projet Celsius estimate, tokens of the task × electricity of writing 1,000 tokens for the size of the model (AI Impact Database, IA-0011) × data centre overhead; range between "all tokens read" and "all written". When the size comes from the price, a neighbouring size would divide the estimate by 4 to 6 or multiply it by as much: the profile of each model specifies this. Benchmark: a quick question to an assistant needs about 0.3 Wh. The quick question in the AI Impact Database.
Continue in the ranking
The full ranking →The hub newsletter
This month's ranking, news on the impact of AI and one figure from the AI Impact Database. One newsletter a month at most, free of charge. Past issues







