[2603.02949] SEALing the Gap: A Reference Framework for LLM Inference Carbon Estimation via Multi-Benchmark Driven Embodiment

Computer Science > Software Engineering arXiv:2603.02949 (cs) [Submitted on 3 Mar 2026] Title:SEALing the Gap: A Reference Framework for LLM Inference Carbon Estimation via Multi-Benchmark Driven Embodiment Authors:Priyavanshi Pathania, Rohit Mehra, Vibhu Saujanya Sharma, Vikrant Kaulgud, Tiffani Nevels, Sanjay Podder, Adam P. Burden View a PDF of the paper titled SEALing the Gap: A Reference Framework for LLM Inference Carbon Estimation via Multi-Benchmark Driven Embodiment, by Priyavanshi Pathania and 6 other authors View PDF HTML (experimental) Abstract:Large Language Models are rapidly gaining traction in software engineering, yet their growing carbon footprint raises pressing sustainability concerns. While training emissions are substantial, inference quickly surpasses them due to the sheer volume of prompts processed. This shift underscores the urgent need for accurate, prompt-level carbon measurement during inference to enable informed, sustainability-focused decision-making. To address the limitations of existing approaches, in this paper, we outline the guiding principles for a novel reference framework for LLM inference carbon estimation that can guide the design of future tools and provide a systematic foundation for advancing sustainability research in this domain. We also introduce SEAL, an early embodiment of these principles that leverages a multi-benchmark-driven approach for per-prompt carbon estimation. Its initial validation shows promising results, posi...