OpenAI’s newly released GPT-5 may use significantly more electricity per response than earlier versions of ChatGPT, according to academic researchers benchmarking AI energy consumption.

First reported by the Guardian, it was revealed that tests undertaken by the University of Rhode Island’s AI lab showed that a medium-length GPT-5 output of around 1,000 tokens can require, on average, more than 18 watt-hours of electricity, with some responses reaching 40 watt-hours.

This is significantly higher than GPT-4o, OpenAI’s prior flagship model.

For reference, 18 watt-hours is around the same as running an incandescent bulb for 18 minutes. Therefore, it was projected that if GPT-5 were to handle 2.5 billion daily queries, as some reports suggest, that level of usage could equal the daily electricity demand of about 1.5 million US homes.

Energy researchers cited in the Guardian report noted that GPT-5 is likely much larger than GPT-4, which itself was believed to be ten times the size of GPT-3. Larger models have consistently shown higher resource needs during both training and inference.

A study released by French AI firm Mistral this summer found a strong correlation between model size and energy consumption, estimating that a model ten times larger will have an environmental footprint roughly an order of magnitude greater for the same output.

While OpenAI has not disclosed GPT-5’s parameter count or detailed energy usage, experts interviewed by the Guardian, which included academics from the University of Illinois, UC Riverside, and the University of Rhode Island, expect its energy demand to be “orders of magnitude” greater than GPT-3.

Last month, OpenAI CEO Sam Altman provided a glimpse into the individual power and water usage of ChatGPT.

"The average query uses about 0.34 watt-hours, about what an oven would use in a little over one second, or a high-efficiency lightbulb would use in a couple of minutes," Altman said in a blog post. "It also uses about 0.000085 gallons of water; roughly one-fifteenth of a teaspoon."

Altman may now have to revise those figures due to the apparent energy demand of OpenAI’s latest model.