{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T05:03:01Z","timestamp":1789448581120,"version":"build-2803163510"},"reference-count":39,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T00:00:00Z","timestamp":1789430400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/501100004728","name":"VIT University","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100004728","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"<jats:p>Deploying large language models in practice means balancing not just answer quality but also the cost and latency of producing each answer, a balance that a single fixed model rarely achieves effectively. To address this issue, two methodologies are adopted: using the best model while keeping the prompt fixed, and changing the prompt according to the query while changing the model. A model has been designed that uses the Prompt\u2013Router Co-Evolutionary Problem (PRCP) and CORAL, which stands for Co-Evolutionary Optimization of Routing and Adaptive Prompts. It considers model selection and prompt patterns as interrelated operations and is structured to improve the performance of both simultaneously. This is achieved by maintaining two groups: one for prompts and the other for routing strategies, allowing a linked interconnection between them. This entire process is guided by multi-objective optimization techniques (NSGA-II\/III) to improve correctness, cost balance, throughput, and latency. The test datasets used for evaluation include MMLU, GSM8K, and Wild Chat, which measure AI model performance. The results show that CORAL is more cost-efficient than existing methods. On GSM8K, at a matched accuracy level, it reduces cost by 31.4 percent relative to RouteLLM. At a matched cost budget, it increases accuracy by up to 4.8 percentage points, again on GSM8K against RouteLLM, and lowers the average per-query latency by 12.2 percent. Taken together, these findings suggest that treating prompt design and model selection as coupled decisions rather than independent steps offers a practical path toward lower-cost, higher-accuracy LLM deployment. The PRCP formulation introduced here provides a reusable framework for studying this coupling in future work.<\/jats:p>","DOI":"10.3389\/frai.2026.1891737","type":"journal-article","created":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T04:27:11Z","timestamp":1789446431000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["CORAL: co-evolutionary optimization of routing and adaptive prompts for cost-efficient LLM deployment"],"prefix":"10.3389","volume":"9","author":[{"given":"Shlok","family":"Goenka","sequence":"first","affiliation":[{"name":"School of Computer Science and Engineering, Vellore Institute of Technology","place":["Vellore, Tamil Nadu, India"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ganesh","family":"Khekare","sequence":"additional","affiliation":[{"name":"School of Computer Science and Engineering, Vellore Institute of Technology","place":["Vellore, Tamil Nadu, India"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Prithu","family":"Adhikari","sequence":"additional","affiliation":[{"name":"School of Computer Science and Engineering, Vellore Institute of Technology","place":["Vellore, Tamil Nadu, India"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Anushree","family":"Tyagi","sequence":"additional","affiliation":[{"name":"School of Computer Science and Engineering, Vellore Institute of Technology","place":["Vellore, Tamil Nadu, India"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Divyanjali","family":"Tyagi","sequence":"additional","affiliation":[{"name":"School of Computer Science and Engineering, Vellore Institute of Technology","place":["Vellore, Tamil Nadu, India"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,9,15]]},"reference":[{"key":"B1","article-title":"\u201cAutoMix: automatically mixing language models,\u201d","volume-title":"Proceedings of the 2024 conference on empirical methods in natural language processing","author":"Aggarwal","year":"2024"},{"key":"B2","article-title":"MOPrompt: multi-objective semantic evolution for prompt optimization","author":"Camara","year":"2025","journal-title":"arXiv"},{"key":"B3","article-title":"FrugalGPT: how to use large language models while reducing cost and improving performance","author":"Chen","year":"2023","journal-title":"arXiv"},{"key":"B4","article-title":"\u201cChatbot arena: an open platform for evaluating LLMs by human preference,\u201d","volume-title":"Proceedings of the 41st international conference on machine learning, ICML","author":"Chiang","year":"2024"},{"key":"B5","article-title":"Training verifiers to solve math word problems","author":"Cobbe","year":"2021","journal-title":"arXiv"},{"key":"B6","doi-asserted-by":"publisher","first-page":"577","DOI":"10.1109\/TEVC.2013.2281535","article-title":"An evolutionary many-objective optimization algorithm using reference-point-based non-dominated sorting approach, Part I: Solving problems with box constraints","volume":"18","author":"Deb","year":"2014","journal-title":"IEEE Trans. Evol. Comput."},{"key":"B7","doi-asserted-by":"publisher","first-page":"182","DOI":"10.1109\/4235.996017","article-title":"A fast and elitist multiobjective genetic algorithm: NSGA-II","volume":"6","author":"Deb","year":"2002","journal-title":"IEEE Trans. Evol. Comput."},{"key":"B8","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2022.emnlp-main.222","article-title":"\u201cRLPrompt: optimizing discrete text prompts with reinforcement learning,\u201d","volume-title":"Proceedings of the 2022 conference on empirical methods in natural language processing","author":"Deng","year":"2022"},{"key":"B9","article-title":"\u201cHybrid LLM: cost-efficient and quality-aware query routing,\u201d","volume-title":"The twelfth international conference on learning representations, ICLR 2024","author":"Ding","year":"2024"},{"key":"B10","article-title":"GraphRouter: a graph-based router for LLM selections","author":"Feng","year":"2024","journal-title":"arXiv"},{"key":"B11","article-title":"PromptBreeder: self-referential self-improvement via prompt evolution","author":"Fernando","year":"2023","journal-title":"arXiv"},{"key":"B12","article-title":"\u201cConnecting large language models with evolutionary algorithms yields powerful prompt optimizers,\u201d","volume-title":"International conference on learning representations, ICLR 2024","author":"Guo","year":"2024"},{"key":"B13","article-title":"\u201cMeasuring massive multitask language understanding,\u201d","volume-title":"Proceedings of the international conference on learning representations, ICLR 2021","author":"Hendrycks","year":"2021"},{"key":"B14","article-title":"RouterBench: a benchmark for multi-LLM routing system","author":"Hu","year":"2024","journal-title":"arXiv"},{"key":"B15","article-title":"Mixtral of experts","author":"Jiang","year":"2024","journal-title":"arXiv"},{"key":"B16","doi-asserted-by":"publisher","first-page":"14165","DOI":"10.18653\/v1\/2023.acl-long.792","article-title":"\u201cLLM-Blender: ensembling large language models with pairwise ranking and generative fusion,\u201d","author":"Jiang","year":"2023","journal-title":"Proceedings of the 61st annual meeting of the association for computational linguistics (ACL 2023)"},{"key":"B17","article-title":"\u201cDSPy: compiling declarative language model calls into self-improving pipelines,\u201d","volume-title":"International conference on learning representations, ICLR 2024","author":"Khattab","year":"2024"},{"key":"B18","doi-asserted-by":"publisher","DOI":"10.14569\/IJACSA.2023.0141262","article-title":"Optimizing network security and performance through the integration of hybrid GAN-RNN models in SDN-based access control and traffic engineering","author":"Khekare","year":"2023","journal-title":"Int. J. Adv. Comput. Sci. Appl."},{"key":"B19","doi-asserted-by":"crossref","DOI":"10.1109\/ICCE50343.2020.9290669","article-title":"\u201cDesign of automatic key finder for search engine optimization in internet of everything,\u201d","volume-title":"2020 IEEE 1st international conference for convergence in engineering (ICCE)","author":"Khekare","year":"2020"},{"key":"B20","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2021.acl-long.353","article-title":"\u201cPrefix-tuning: optimizing continuous prompts for generation,\u201d","volume-title":"Proceedings of the 59th annual meeting of the association for computational linguistics","author":"Li","year":"2021"},{"key":"B21","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2022.acl-short.8","article-title":"\u201cP-tuning: prompt tuning can be comparable to fine-tuning across scales and tasks,\u201d","volume-title":"Proceedings of the 60th annual meeting of the association for computational linguistics","author":"Liu","year":"2022"},{"key":"B22","article-title":"Routing to the expert: efficient reward-guided ensemble of large language models","author":"Lu","year":"2023","journal-title":"arXiv"},{"key":"B23","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2022.acl-long.556","article-title":"\u201cFantastically ordered prompts and where to find them: overcoming few-shot prompt order sensitivity,\u201d","volume-title":"Proceedings of the 60th annual meeting of the association for computational linguistics","author":"Lu","year":"2022"},{"key":"B24","article-title":"\u201cRouteLLM: learning to route LLMs with preference data,\u201d","volume-title":"International conference on learning representations, ICLR 2025","author":"Ong","year":"2024"},{"key":"B25","article-title":"GPT-4 technical report","year":"2024","journal-title":"arXiv"},{"key":"B26","doi-asserted-by":"crossref","DOI":"10.1007\/3-540-58484-6_269","article-title":"\u201cA cooperative coevolutionary approach to function optimization,\u201d","volume-title":"Parallel Problem Solving from Nature, PPSN III, Lecture Notes in Computer Science","author":"Potter","year":"1994"},{"key":"B27","doi-asserted-by":"publisher","first-page":"1573","DOI":"10.1007\/s11227-025-08034-8","article-title":"A data-augmented model routing framework for efficient LLM deployment in edge-cloud environments","volume":"81","author":"Pozi","year":"2025","journal-title":"J. Supercomput."},{"key":"B28","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2023.emnlp-main.494","article-title":"\u201cAutomatic prompt optimization with \u2018gradient descent' and beam search,\u201d","volume-title":"Proceedings of the 2023 conference on empirical methods in natural language processing","author":"Pryzant","year":"2023"},{"key":"B29","doi-asserted-by":"publisher","first-page":"24721","DOI":"10.1038\/s41598-026-64427-6","article-title":"Explainability-driven adaptive cyber deception control system for autonomous network defense","volume":"16","author":"Roy","year":"2026","journal-title":"Sci. Rep."},{"key":"B30","article-title":"Large language model routing with benchmark datasets","author":"Shnitzer","year":"2023","journal-title":"arXiv"},{"key":"B31","doi-asserted-by":"publisher","first-page":"11384","DOI":"10.1038\/s41598-026-43692-5","article-title":"Optimizing charge discharge cycles using QPPONet-enabled hybrid learning framework for energy management and safety in electric vehicles","volume":"16","author":"Sujan Kumar","year":"2026","journal-title":"Sci. Rep."},{"key":"B32","doi-asserted-by":"crossref","DOI":"10.52202\/068431-1800","article-title":"\u201cChain-of-thought prompting elicits reasoning in large language models,\u201d","volume-title":"Advances in neural information processing systems, NeurIPS 2022","author":"Wei","year":"2022"},{"key":"B33","article-title":"\u201cLarge language models as optimizers,\u201d","volume-title":"International conference on learning representations, ICLR 2024","author":"Yang","year":"2024"},{"key":"B34","article-title":"TextGrad: automatic \u2018differentiation' via text","author":"Yuksekgonul","year":"2024","journal-title":"arXiv"},{"key":"B35","article-title":"\u201cIRT-Router: effective and interpretable multiLLM routing using item response theory,\u201d","volume-title":"Proceedings of the 63rd annual meeting of the association for computational linguistics (ACL 2025)","author":"Zhang","year":"2025"},{"key":"B36","article-title":"\u201cWildChat: 1M ChatGPT interaction logs in the wild,\u201d","volume-title":"International conference on learning representations, ICLR 2024","author":"Zhao","year":"2024"},{"key":"B37","article-title":"\u201cCalibrate before use: Improving few-shot performance of language models,\u201d","volume-title":"Proceedings of the 38th international conference on machine learning, ICML 2021","author":"Zhao","year":"2021"},{"key":"B38","doi-asserted-by":"crossref","DOI":"10.52202\/075280-2020","article-title":"\u201cJudging LLM-as-a-judge with MT-bench and chatbot arena,\u201d","volume-title":"Advances in neural information processing systems (NeurIPS 2023)","author":"Zheng","year":"2023"},{"key":"B39","article-title":"Large language models are human-level prompt engineers","author":"Zhou","year":"2022","journal-title":"arXiv"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1891737\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,15]],"date-time":"2026-09-15T04:27:18Z","timestamp":1789446438000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1891737\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,15]]},"references-count":39,"alternative-id":["10.3389\/frai.2026.1891737"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1891737","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,15]]},"article-number":"1891737"}}