RoBoN: Routed Online Best-of-n for Test-Time Scaling with Multiple LLMs

• URL: http://arxiv.org/abs/2512.05542v1
• Date: Fri, 05 Dec 2025 08:55:39 GMT
• Title: RoBoN: Routed Online Best-of-n for Test-Time Scaling with Multiple LLMs
• Authors: Jonathan Geuter, Gregor Kornhardt,
• Abstract summary: Routed Online Best-of-$n$ is a sequential multi-LLM alternative to the prevailing single-model best-of-$n$.<n>Our results indicate that diversity across models can be exploited at inference to improve best-of-$n$ performance over any constituent model alone.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Best-of-$n$ is a widely used test-time scaling approach for LLM inference. Yet despite evidence that LLMs exhibit complementary strengths across tasks, traditionally best-of-$n$ relies on a single model to generate responses. We propose RoBoN (Routed Online Best-of-$n$), a sequential multi-LLM alternative to the prevailing single-model best-of-$n$. Given a suite of models $\{m_i\}_{i=1}^M$, RoBoN sequentially routes generations one-by-one across models, based on scores computed using a reward model and an agreement signal on the predicted responses. This online routing requires no additional training, keeps compute parity, and works with any plug-in reward model. Across reasoning benchmarks (MATH500, OlympiadBench, MinervaMath, GSM8K, MMLU), RoBoN consistently outperforms standard best-of-$n$ applied to each individual model for larger $n$, with gains of up to 3.4\% in absolute accuracy, and also improves over a uniform multi-model portfolio baseline. Our results indicate that diversity across models can be exploited at inference to improve best-of-$n$ performance over any constituent model alone, providing a simple, training-free path to test-time scaling with multiple LLMs.

Related papers

• MMR-Bench: A Comprehensive Benchmark for Multimodal LLM Routing [41.77627136743721]
  In practical deployments, workloads span lightweight OCR to complex multimodal reasoning.<n>routing is nontrivial due to modality fusion, wide variation in computational cost across models, and the absence of a standardized, budget-aware evaluation.<n>We present MMR-Bench, a unified benchmark that isolates the multimodal routing problem and enables comparison under fixed candidate sets and cost models.
  arXiv  Detail & Related papers  (2026-01-25T12:44:14Z)
• DISCO: Diversifying Sample Condensation for Efficient Model Evaluation [59.01400190971061]
  Costly evaluation reduces inclusivity, slows the cycle of innovation, and worsens environmental impact.<n>We argue that promoting diversity among samples is not essential; what matters is to select samples thatmaximise diversity in model responses.<n>Our method, $textbfDiversifying Sample Condensation (DISCO)$, selects the top-k samples with the greatest model disagreements.
  arXiv  Detail & Related papers  (2025-10-09T08:53:59Z)
• OptMerge: Unifying Multimodal LLM Capabilities and Modalities via Model Merging [124.91183814854126]
  Model merging seeks to combine multiple expert models into a single model.<n>We introduce a benchmark for model merging research that clearly divides the tasks for MLLM training and evaluation.<n>We find that model merging offers a promising way for building improved MLLMs without requiring training data.
  arXiv  Detail & Related papers  (2025-05-26T12:23:14Z)
• Capability Instruction Tuning: A New Paradigm for Dynamic LLM Routing [64.38277118982698]
  Large Language Models (LLMs) have demonstrated human-like instruction-following abilities.<n>In this work, we explore how to route the best-performing LLM for each instruction to achieve better overall performance.<n>We develop a new paradigm, constructing capability instructions with model capability representation, user instruction, and performance inquiry prompts to assess the performance.
  arXiv  Detail & Related papers  (2025-02-24T16:10:53Z)
• Transfer Q Star: Principled Decoding for LLM Alignment [105.89114186982972]
  Transfer $Q*$ estimates the optimal value function for a target reward $r$ through a baseline model. Our approach significantly reduces the sub-optimality gap observed in prior SoTA methods.
  arXiv  Detail & Related papers  (2024-05-30T21:36:12Z)
• One-shot Active Learning Based on Lewis Weight Sampling for Multiple Deep Models [39.582100727546816]
  Active learning (AL) for multiple target models aims to reduce labeled data querying while effectively training multiple models concurrently. Existing AL algorithms often rely on iterative model training, which can be computationally expensive. We propose a one-shot AL method to address this challenge, which performs all label queries without repeated model training.
  arXiv  Detail & Related papers  (2024-05-23T02:48:16Z)
• Large Language Model Routing with Benchmark Datasets [40.42044096089315]
  No single model typically achieves the best accuracy in all tasks and use cases. We propose a new formulation for the problem, in which benchmark datasets are repurposed to learn a "router" model for this selection. We show that this problem can be reduced to a collection of binary classification tasks.
  arXiv  Detail & Related papers  (2023-09-27T17:08:40Z)
• Improving Robustness and Generality of NLP Models Using Disentangled Representations [62.08794500431367]
  Supervised neural networks first map an input $x$ to a single representation $z$, and then map $z$ to the output label $y$. We present methods to improve robustness and generality of NLP models from the standpoint of disentangled representation learning. We show that models trained with the proposed criteria provide better robustness and domain adaptation ability in a wide range of supervised learning tasks.
  arXiv  Detail & Related papers  (2020-09-21T02:48:46Z)
• Model-Based Multi-Agent RL in Zero-Sum Markov Games with Near-Optimal Sample Complexity [67.02490430380415]
  We show that model-based MARL achieves a sample complexity of $tilde O(|S||B|(gamma)-3epsilon-2)$ for finding the Nash equilibrium (NE) value up to some $epsilon$ error. We also show that such a sample bound is minimax-optimal (up to logarithmic factors) if the algorithm is reward-agnostic, where the algorithm queries state transition samples without reward knowledge.
  arXiv  Detail & Related papers  (2020-07-15T03:25:24Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.