R2-Router: A New Paradigm for LLM Routing with Reasoning

TL;DR

R2-Router introduces a novel LLM routing approach that jointly optimizes LLM choice and output length budget, enabling cost-effective high-quality responses and surpassing existing methods in efficiency.

Contribution

It presents R2-Router, a new paradigm that considers output length as a controllable variable for improved LLM routing, and introduces R2-Bench, a dataset for evaluating this approach.

Findings

R2-Router achieves 4-5x lower cost than existing routers.

Joint selection of LLM and length budget improves performance.

R2-Bench effectively captures LLM behavior across budgets.

Abstract

As LLMs proliferate with diverse capabilities and costs, LLM routing has emerged by learning to predict each LLM's quality and cost for a given query, then selecting the one with high quality and low cost. However, existing routers implicitly assume a single fixed quality and cost per LLM for each query, ignoring that the same LLM's quality varies with its output length. This causes routers to exclude powerful LLMs when their estimated cost exceeds the budget, missing the opportunity that these LLMs could still deliver high quality at reduced cost with shorter outputs. To address this, we introduce R2-Router, which treats output length budget as a controllable variable and jointly selects the best LLM and length budget, enforcing the budget via length-constrained instructions. This enables R2-Router to discover that a powerful LLM with constrained output can outperform a weaker LLM at comparable cost-efficient configurations invisible to prior methods. Together with the router framework, we construct R2-Bench, the first routing dataset capturing LLM behavior across diverse output length budgets. Experiments show that R2-Router achieves state-of-the-art performance at 4-5x lower cost compared with existing routers. This work opens a new direction: routing as reasoning, where routers evolve from reactive selectors to deliberate reasoners that explore which LLM to use and at what cost budget.