The Hidden Cost of Thinking: Energy Use and Environmental Impact of LMs Beyond Pretraining

• URL: http://arxiv.org/abs/2605.01158v1
• Date: Fri, 01 May 2026 23:24:23 GMT
• Title: The Hidden Cost of Thinking: Energy Use and Environmental Impact of LMs Beyond Pretraining
• Abstract summary: Olmo 3 is a family of 7 billion and 32 billion parameter models in both instruction-following and reasoning variants.<n>We estimate our model development process consumed 12.3 GWh of datacenter energy, emitted 4,251 tCO2eq, and consumed 15,887 kL of water, with water consumption driven entirely by power generation infrastructure rather than data center cooling.
• Score: 59.27959962602072
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Modern language model development extends far beyond pretraining, yet environmental reporting remains narrowly focused on the cost of training a single final model. In this work, we provide the first detailed breakdown of the environmental impact of a full model development pipeline, from pretraining through supervised fine-tuning, preference optimization, and reinforcement learning, for Olmo 3, a family of 7 billion and 32 billion parameter models in both instruction-following and reasoning variants. We find that reasoning models are 17x more expensive to post-train than their instruction-tuned counterparts in terms of datacenter energy, driven by reinforcement learning rollout generation. Development costs (including experimentation, failed runs, and ablations) account for 82.2% of total compute, a roughly 65% increase over the ~50% reported for pretraining-focused pipelines in prior work. In total, we estimate our model development process consumed ~12.3 GWh of datacenter energy, emitted 4,251 tCO2eq, and consumed 15,887 kL of water, with water consumption driven entirely by power generation infrastructure rather than data center cooling. These costs, which are almost entirely unreported by model developers, are growing rapidly as post-training pipelines become more complex, and must be accounted for in environmental reporting standards and by the research community working to reduce AI's environmental impact.

Related papers

• Towards Green AI: Decoding the Energy of LLM Inference in Software Development [46.879983975894135]
  AI-assisted tools are increasingly integrated into software development, but their reliance on large language models (LLMs) introduces substantial computational and energy costs.<n>We conduct a phase-level analysis of LLM inference energy consumption, distinguishing between the (1) prefill, where the model processes the input and builds internal representations, and (2) decoding, where output tokens are generated using the stored state.
  arXiv  Detail & Related papers  (2026-02-05T14:38:19Z)
• Recycling Pretrained Checkpoints: Orthogonal Growth of Mixture-of-Experts for Efficient Large Language Model Pre-Training [70.60554423630803]
  We propose to recycle pretrained checkpoints by expanding their parameter counts and continuing training.<n>We scale our approach to models with 70B parameters and over 1T training tokens, achieving 10.66% accuracy gain over training from scratch.
  arXiv  Detail & Related papers  (2025-10-09T09:45:45Z)
• How Hungry is AI? Benchmarking Energy, Water, and Carbon Footprint of LLM Inference [0.0]
  This paper introduces a novel infrastructure-aware benchmarking framework for quantifying the environmental footprint of AI inference across 30 state-of-the-art models as deployed in commercial data centers.<n>Our results show that o3 and DeepSeek-R1 emerge as the most energy-intensive models, consuming over 33 Wh per long prompt, more than 70 times the consumption of GPT-4.1 nano, and that Claude-3.7 Sonnet ranks highest in eco-efficiency.<n>These findings illustrate a growing paradox: Although AI is becoming cheaper and faster, its global adoption drives disproportionate resource consumption.
  arXiv  Detail & Related papers  (2025-05-14T17:47:00Z)
• Every FLOP Counts: Scaling a 300B Mixture-of-Experts LING LLM without Premium GPUs [96.68469559192846]
  We present two differently sized MoE large language models (LLMs)<n>Ling-Lite contains 16.8 billion parameters with 2.75 billion activated parameters, while Ling-Plus boasts 290 billion parameters with 28.8 billion activated parameters.<n>We propose innovative methods for (1) optimization of model architecture and training processes, (2) refinement of training anomaly handling, and (3) enhancement of model evaluation efficiency.
  arXiv  Detail & Related papers  (2025-03-07T04:43:39Z)
• Holistically Evaluating the Environmental Impact of Creating Language Models [26.846990296567267]
  We estimate the real-world environmental impact of developing a series of language models, ranging from 20 million to 13 billion active parameters, trained on up to 5.6 trillion tokens each.<n>We find that our series of models released 493 metric tons of carbon emissions, equivalent to powering about 98 homes in the United States for one year.<n>We also find that power usage throughout training is not consistent, fluctuating between 15% and 85% of our hardware's maximum power draw.
  arXiv  Detail & Related papers  (2025-03-03T22:16:15Z)
• Software Frugality in an Accelerating World: the Case of Continuous Integration [2.73028688816111]
  We conduct the first large-scale analysis of the energy footprint of Continuous Integration pipelines implemented with GitHub. We observe that the average unitary energy cost of a pipeline is relatively low, at 10 Wh. When evaluating CO2 emissions based on regional Wh-to-CO2 estimates, we observe that the average aggregated CO2 emissions are significant, averaging 10.5 kg.
  arXiv  Detail & Related papers  (2024-10-21T09:29:50Z)
• Reporting and Analysing the Environmental Impact of Language Models on the Example of Commonsense Question Answering with External Knowledge [7.419725234099729]
  ChatGPT sparked social interest in Large Language Models (LLMs) LLMs demand substantial computational resources and are very costly to train, both financially and environmentally. In this study, we infused T5 LLM with external knowledge and fine-tuned the model for Question-Answering task.
  arXiv  Detail & Related papers  (2024-07-24T16:16:16Z)
• Estimating the Carbon Footprint of BLOOM, a 176B Parameter Language Model [72.65502770895417]
  We quantify the carbon footprint of BLOOM, a 176-billion parameter language model, across its life cycle. We estimate that BLOOM's final training emitted approximately 24.7 tonnes ofcarboneqif we consider only the dynamic power consumption. We conclude with a discussion regarding the difficulty of precisely estimating the carbon footprint of machine learning models.
  arXiv  Detail & Related papers  (2022-11-03T17:13:48Z)
• Low Emission Building Control with Zero-Shot Reinforcement Learning [70.70479436076238]
  Control via Reinforcement Learning (RL) has been shown to significantly improve building energy efficiency. We show it is possible to obtain emission-reducing policies without a priori--a paradigm we call zero-shot building control.
  arXiv  Detail & Related papers  (2022-08-12T17:13:25Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.