Thinking AI models like ChatGPT emit '50 times more CO2' but still give wrong answers

Daily Record

24-06-2025

The more an AI service thinks, the more carbon it emits.
Artificial Intelligence is a tool being used by millions of people the world over. AI is when computer systems perform tasks that typically require human intelligence, such as learning, problem-solving, and decision-making.
From homeowners asking ChatGPT for renovation advice, to the software revealing what Scottish homes could look like in the next 25 years, engaging with AI can be helpful and eye-opening, but can also come with serious risks.
‌
A recent study from MIT found that using ChatGPT for essay writing can negatively impact cognitive engagemen t and memory recall, compared to those who wrote purely from their own brain.
‌
But it's not just the personal impact AI can have, it can also damage the environment. Another study analysing different types of AI found there was a marked difference in CO2 output depending on the model.
A query typed into a large language model (LLM), such as ChatGPT, requires energy and produces more CO2 emissions. Emissions, however, depend on the model, the subject matter, and the user.
Researchers compared 14 models and found that complex answers cause more emissions than simple answers. Meanwhile, models that provide more accurate answers also produce more emissions.
Wondering how asking AI a question produces CO2 emissions? Well, no matter which questions we ask an AI, the model will come up with an answer, the researchers in Germany explained.
To produce this information - regardless of whether that answer is correct or not - the model uses tokens. Tokens are words or parts of words that are converted into a string of numbers that can be processed by the LLM.
‌
This conversion, as well as other computing processes, produce CO2 emissions. Many users, however, are unaware of the substantial carbon footprint associated with these technologies.
With that in mind, researchers measured and compared CO2 emissions of different, already trained, LLMs using a set of standardised questions.
"The environmental impact of questioning trained LLMs is strongly determined by their reasoning approach," explained first author Maximilian Dauner.
‌
"Explicit reasoning processes significantly drive up energy consumption and carbon emissions. We found that reasoning-enabled models produced up to 50 times more CO2 emissions than concise response models."
'Thinking' AI causes the most emissions. Reasoning models, on average, created 543.5 'thinking' tokens per question, whereas concise models required just 37.7 tokens per question.
‌
Thinking tokens are additional tokens that reasoning LLMs generate before producing an answer. A higher token footprint always means higher CO2 emissions.
It doesn't, however, mean the resulting answers are more correct. This is because elaborate detail does not always equal correctness.
‌
Subject matter also resulted in significantly different levels of CO2 emissions. Questions that required lengthy reasoning processes, for example abstract algebra or philosophy, led to up to six times higher emissions than more straightforward subjects, like high school history.
The most accurate model was the Cogito model with 70 billion parameters, reaching 84.9 per cent accuracy. The model produced three times more CO2 emissions than similar sized models that generated concise answers.
All is not lost, though. If you are a tech enthusiast, but also climate-conscious, you can, to an extent, control the amount of CO2 emissions caused by AI by adjusting your personal use of the technology, the researchers said.
‌
"Users can significantly reduce emissions by prompting AI to generate concise answers or limiting the use of high-capacity models to tasks that genuinely require that power," Dauner pointed out.
Choice of model can make a big difference in CO2 emissions. For example, having DeepSeek R1 answer 600,000 questions would create CO2 emissions equal to a round-trip flight from London to New York.
Meanwhile, OpenAI's ChatGPT consumes 500 ml of water for every five to 50 prompts it answers, according to Shaolei Ren, a researcher at the University of California, Riverside.
"If users know the exact CO2 cost of their AI-generated outputs, such as casually turning themselves into an action figure, they might be more selective about when and how they use these technologies," Dauner said.
Join the Daily Record WhatsApp community!