Slm vs Llm: A comparative study on cost, efficiency and enterprise adoption

This paper compared small and large language models on efficiency, cost, and adoption. SLMs require much less memory and hardware, can be served locally, and, under the stated assumptions, repay their upfront cost within roughly a year at about one million requests per month, sooner at higher volume or when compared with premium APIs. LLMs retain their advantage for complex reasoning and novel problems. Adoption follows technical maturity, data-governance needs, and workload volume. The results indicate that the question is not which model class is better but which is right for each task. A routing-based hybrid, using SLMs for frequent well-defined work and LLMs for hard cases, offers a practical way to balance quality, cost, and control.

1.      A. Vaswani et al., “Attention Is All You Need,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 30, 2017. arXiv:1706.03762. 2.      T. Brown et al., “Language Models are Few-Shot Learners,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 33, 2020. arXiv:2005.14165. 3.      H. Touvron et al., “Llama 2: Open Foundation and Fine-Tuned Chat Models,” arXiv:2307.09288, 2023. 4.      A. Q. Jiang et al., “Mistral 7B,” arXiv:2310.06825, 2023. 5.      M. Abdin et al., “Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone,” arXiv:2404.14219, 2024. 6.      G. Hinton, O. Vinyals, and J. Dean, “Distilling the Knowledge in a Neural Network,” arXiv:1503.02531, 2015. 7.      T. Dettmers, A. Pagnoni, A. Holtzman, and L. Zettlemoyer, “QLoRA: Efficient Finetuning of Quantized LLMs,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 36, 2023. arXiv:2305.14314. 8.      W. Kwon et al., “Efficient Memory Management for Large Language Model Serving with PagedAttention,” in Proc. ACM Symposium on Operating Systems Principles (SOSP), 2023. arXiv:2309.06180. 9.      L. Chen, M. Zaharia, and J. Zou, “FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance,” arXiv:2305.05176, 2023. 10.   P. Belcak et al., “Small Language Models are the Future of Agentic AI,” arXiv:2506.02153, 2025.