Cheap or free access to AI models, not only paying the latest models, but also improves with the latest firm available for all users. However, this entry comes with an expense: the environment.
In a new study, German researchers tested 14 large language models (LLMS) of various sizes, such as Meta, Alibaba and others. Each model answered 1,000 difficult academic questions covering topics for advanced mathematics from the history of world history. Tests fled to a strong, energy intensive NVIDIA A100 GPU, using a special framework to accurately measure electricity consumption. This information was then turned into carbon dioxide equivalent emissions by comparing the environmental effects of each model.
Researchers have determined that many LLMs are stronger than necessary for daily surveys. The smaller, less energy-hungry models can also answer many actual questions. A single urgent carbon and waterprints change sharply depending on the model size and task type. As they have created more verses, the models for “think aloud” require “think aloud”, especially polluting substantiation.
One model, COGITO, Accuracy Schedule – answered approximately 85% of the questions, but three times more emissions, or users of similar sized models, three times more emissions from similarly visible models. (COGITO did not respond to a survey.) “World War II World War II says Maximilian Dauner, a researcher and one of the teaching authors at the University of Applied Sciences.
Results stressed the balance between accuracy and waste. The least pollutant model has been tested, Qwen 7B responded to only one of the three questions, but a total of 27.7 grams of carbon dioxide equivalent. On the contrary, Deepseek’s R1 70B thought model responded to 10 questions for more than 70 times more than 70 times for the same workload.
The type of question affects the environment. Algebra or philosophy, a high school student removes waste from six times higher than how much you create homework.
“Companies must be more transparent about real waste and water consumers,” Dauner says. But at the same time, users should be more aware of and should be more mild.