From Usage to Cost: A Behavior-Aware Web Tool for Cost Estimation and Simulation of LLM-Based Applications
Resumo
Large language model (LLM) applications introduce usage-dependent operational costs that are affected not only by model pricing but also by how users interact with the system. Conventional estimates based on aggregate request volumes or average token consumption may overlook important workload heterogeneity. This paper presents a Web-based decision-support tool that uses historical application workloads to characterize differences across user profiles and request types and uses this behavior to explore prospective cost scenarios. In addition to analyzing current costs at population, profile, and request-type levels, the tool allows analysts to vary population size, profile composition, workload statistics, and LLM pricing independently. The tool was developed using historical data from Ana Health, a Brazilian digital primary-care provider. In the analyzed snapshot, a profile containing 18.2% of users generated 31.9% of requests and accounted for 32.9% of estimated cost, illustrating why aggregate averages may hide relevant differences. By connecting historical behavior with configurable future scenarios, the tool supports cost planning and assessment of the economic impact of workload and population changes.
Referências
Vahid Farajijobehdar, İlknur Köseoğlu Sarı, Nazım Kemal Üre, and Engin Zeydan. 2026. Tokalator: A Context Engineering Toolkit for Artificial Intelligence Coding Assistants. arXiv:2604.08290 [cs.SE] [link]
Anderson A. Ferreira, Leonardo Rocha,Washington Cunha, Ana Cláudia Machado, João Marcos Campos, Gabriel Jallais, Adriana C. F. Viana, Elisa Tuler, Iago Araújo, Víctor Macul, Olívio Souza Neto, Antônio Pereira de Souza Júnior, Giordano de Pinho Souza, Joice Marques Pallone, Mariana Aparecida Dumbá Soares,Welton Augusto Santos, and Marcos André Gonçalves. 2025. A Comprehensive Qualitative Analysis of Patient Dialogue Summarization Using Large Language Models Applied to Noisy, Informal, Non-English Real-World Data. Scientific Reports 15, 1 (2025), 31660. DOI: 10.1038/s41598-025-13560-9
Malgorzata Lazuka, Andreea Anghel, and Thomas Parnell. 2024. LLM-Pilot: Characterize and Optimize Performance of your LLM Inference Services. In Proceedings of the International Conference for High Performance Computing, Networking, Storage, and Analysis (Atlanta, GA, USA) (SC ’24). IEEE Press, Article 16, 18 pages. DOI: 10.1109/SC41406.2024.00022
Isaac Ong, Amjad Almahairi, Vincent Wu, Wei-Lin Chiang, Tianhao Wu, Joseph E. Gonzalez, M. Waleed Kadous, and Ion Stoica. 2024. RouteLLM: Learning to Route LLMs with Preference Data. arXiv preprint arXiv:2406.18665 (2024). DOI: 10.48550/arXiv.2406.18665
Zhehao Zhang, Ryan A. Rossi, Branislav Kveton, Yijia Shao, Diyi Yang, Hamed Zamani, Franck Dernoncourt, Joe Barrow, Tong Yu, Sungchul Kim, Ruiyi Zhang, Jiuxiang Gu, Tyler Derr, Hongjie Chen, Junda Wu, Xiang Chen, Zichao Wang, Subrata Mitra, Nedim Lipka, Nesreen Ahmed, and YuWang. 2024. Personalization of Large Language Models: A Survey. arXiv preprint arXiv:2411.00027 (2024). DOI: 10.48550/arXiv.2411.00027
