The SIMPLE Law for Energy Cost of LLM Inference

About the SIMPLE Law

The SIMPLE Law is an architecture-aware analytical model that predicts the energy consumed per generated token (E_tok) during LLM inference from a model's architecture (d, L, n_q, n_kv) and its input and output token counts (τ_in, τ_out). This demo plots the SIMPLE Law's prediction and the share of each of its terms. Select one of the models validated in the paper, or load the architecture of any dense decoder-only Transformer (up to approximately 20B parameters) from Hugging Face.

The SIMPLE Law was fitted using this dataset.

For the complete methodology, experimental validation, and citation information, see the paper: The SIMPLE Law for Energy Cost of LLM Inference.

Authors

• Hiari Pizzini Cavagna^a
• Andrea Proia^a
• Daniele Cesarini^b
• Zeynep Kiziltan^a
• Andrea Bartolini^a

• ^a University of Bologna, Bologna, Italy
• ^b CINECA, Casalecchio di Reno, Italy

Citation

@misc{cavagna2026simplelawenergycost,
    title={The SIMPLE Law for Energy Cost of LLM Inference},
    author={Hiari Pizzini Cavagna and Andrea Proia and Daniele Cesarini and Zeynep Kiziltan and Andrea Bartolini},
    year={2026},
    eprint={2611.03330},
    archivePrefix={arXiv},
    primaryClass={cs.AI},
    url={https://arxiv.org/abs/2611.03330},
}

Web demo developed with the assistance of Claude (Anthropic), an AI assistant.