arXiv:2611.25976v1 Announce Type: cross Abstract: We tested whether energy estimates of open-weight large language models (LLMs) scale proportionally with an exactly stated food weight, and whether computing energy density times weight removes the error.