GPT-5 Breakthrough Comes at Steep Energy Cost, Experts Warn

In mid-2023, OpenAI released its latest language model, GPT-5, which powers the popular ChatGPT chatbot. While the new model offers breakthrough capabilities, including the ability to create websites, answer PhD-level science questions, and reason through complex problems, experts warn that these advancements come at a steep cost in terms of energy consumption.

According to researchers, a response from GPT-5 can use up to 40 watt-hours of electricity, which is several times more than the 2 watt-hours required for a similar response from the previous version, GPT-4. On average, GPT-5 consumes just over 18 watt-hours per medium-length response, a figure higher than all other models benchmarked by the researchers, except for OpenAI's own o3 reasoning model and R1 from the Chinese AI firm Deepseek.

These energy demands are significant, with the total consumption of GPT-5 potentially reaching the daily electricity demand of 1.5 million US homes. Experts attribute this increase to the model's larger size, which is believed to be significantly larger than GPT-4, which was already 10 times the size of GPT-3.

While GPT-5 may be deployed on more efficient hardware and use a "mixture-of-experts" architecture to reduce energy consumption, the model's reasoning capabilities and ability to work with video and images likely contribute to its substantial energy footprint. Researchers call on OpenAI and other AI developers to be more transparent about the environmental impact of their increasingly powerful models.