GPT-5 energy use raises questions about power needs

Experts benchmark AI energy use warn that GPT-5 larger size and multimodal features drive higher power demands.

GPT-5 energy use exceeds earlier models

Researchers at the University of Rhode Island measured GPT-5 energy use. They report an average draw of about 18 watt hours for a medium length response and as high as 40 watt hours for roughly 1000 tokens. The results indicate GPT-5 uses more energy than earlier models, a pattern expected as models grow larger.

Experts say the energy cost depends on the architecture; GPT-5 uses a mixture of experts to limit activation, which can reduce usage on some queries. But the model also handles reasoning and multimodal tasks, which tends to increase computation time and energy. OpenAI has not published official energy figures for GPT-5, prompting calls for clearer disclosure and for measuring environmental impact.

The energy cost of AI scales with size and capability. GPT-5 embodies a prominent tension between ambition and cost. The industry will need credible benchmarks and transparent reporting to avoid eroding public trust.

There is a broader policy conversation ahead. Investors and regulators may push for standardized energy metrics, which could influence how future models are designed and deployed. As the field grows, the environmental footprint will shape both public perception and long term strategy for AI labs.

Highlights

As AI grows, clear accounting of energy cost becomes a standard part of responsible innovation.