Reasoning Is Not Free: Robust Adaptive Cost-Efficient Routing for LLM-as-a-Judge

Wenbo Zhang, Lijinghua Zhang, Liner Xiang, Hengrui Cai
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:160731-160755, 2026.

Abstract

Reasoning-capable large language models (LLMs) have recently been adopted as automated judges, but their benefits and costs in LLM-as-a-Judge settings remain unclear. Through controlled comparisons between reasoning and non-reasoning judges, we show that explicit reasoning substantially improves judgment accuracy on tasks requiring structured verification (e.g., math and coding), while offering limited or even negative gains on simpler evaluations and incurring significantly higher computational cost. These findings motivate that reasoning should be used selectively rather than universally, with awareness of possible distribution shift. We propose a Robust Adaptive Cost-Efficient Routing (RACER), which dynamically selects between reasoning and non-reasoning judges under a fixed budget by formulating routing as a constrained distributionally robust optimization problem. RACER explicitly accounts for distribution shift via a KL-divergence uncertainty set, admits an efficient primal–dual algorithm, and enjoys theoretical guarantees including uniqueness of the optimal policy and linear convergence. Extensive experiments show that RACER achieves superior accuracy–cost trade-offs under distribution shift.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-zhang26iv, title = {Reasoning Is Not Free: Robust Adaptive Cost-Efficient Routing for {LLM}-as-a-Judge}, author = {Zhang, Wenbo and Zhang, Lijinghua and Xiang, Liner and Cai, Hengrui}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {160731--160755}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/zhang26iv/zhang26iv.pdf}, url = {https://proceedings.mlr.press/v306/zhang26iv.html}, abstract = {Reasoning-capable large language models (LLMs) have recently been adopted as automated judges, but their benefits and costs in LLM-as-a-Judge settings remain unclear. Through controlled comparisons between reasoning and non-reasoning judges, we show that explicit reasoning substantially improves judgment accuracy on tasks requiring structured verification (e.g., math and coding), while offering limited or even negative gains on simpler evaluations and incurring significantly higher computational cost. These findings motivate that reasoning should be used selectively rather than universally, with awareness of possible distribution shift. We propose a Robust Adaptive Cost-Efficient Routing (RACER), which dynamically selects between reasoning and non-reasoning judges under a fixed budget by formulating routing as a constrained distributionally robust optimization problem. RACER explicitly accounts for distribution shift via a KL-divergence uncertainty set, admits an efficient primal–dual algorithm, and enjoys theoretical guarantees including uniqueness of the optimal policy and linear convergence. Extensive experiments show that RACER achieves superior accuracy–cost trade-offs under distribution shift.} }
Endnote
%0 Conference Paper %T Reasoning Is Not Free: Robust Adaptive Cost-Efficient Routing for LLM-as-a-Judge %A Wenbo Zhang %A Lijinghua Zhang %A Liner Xiang %A Hengrui Cai %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-zhang26iv %I PMLR %P 160731--160755 %U https://proceedings.mlr.press/v306/zhang26iv.html %V 306 %X Reasoning-capable large language models (LLMs) have recently been adopted as automated judges, but their benefits and costs in LLM-as-a-Judge settings remain unclear. Through controlled comparisons between reasoning and non-reasoning judges, we show that explicit reasoning substantially improves judgment accuracy on tasks requiring structured verification (e.g., math and coding), while offering limited or even negative gains on simpler evaluations and incurring significantly higher computational cost. These findings motivate that reasoning should be used selectively rather than universally, with awareness of possible distribution shift. We propose a Robust Adaptive Cost-Efficient Routing (RACER), which dynamically selects between reasoning and non-reasoning judges under a fixed budget by formulating routing as a constrained distributionally robust optimization problem. RACER explicitly accounts for distribution shift via a KL-divergence uncertainty set, admits an efficient primal–dual algorithm, and enjoys theoretical guarantees including uniqueness of the optimal policy and linear convergence. Extensive experiments show that RACER achieves superior accuracy–cost trade-offs under distribution shift.
APA
Zhang, W., Zhang, L., Xiang, L. & Cai, H.. (2026). Reasoning Is Not Free: Robust Adaptive Cost-Efficient Routing for LLM-as-a-Judge. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:160731-160755 Available from https://proceedings.mlr.press/v306/zhang26iv.html.

Related Material