When Strong Reasoners Join the Pool: Small Union Premiums and Costly Runaway Generations in LLM Routing
cs.LG, cs.CL
Submitted: 2026-11-03
Updated: 2026-11-03
Comments: Technical report. 10 pages, 7 figures, 14 tables. Companion to arXiv:2607.03436. Data and code: https://doi.org/10.5281/zenodo.23724735
Code: https://github.com/huggingface/Math-Verify
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Sources
- What Does a Routing Oracle Measure Under Stochastic Decoding? Coupling, Scorer Choice, and Single-Commit Ceilings
- FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance
- RouteLLM: Learning to Route LLMs with Preference Data
- RouterBench: A Benchmark for Multi-LLM Routing System
- The Llama 3 Herd of Models
- NVIDIA Nemotron Nano 2: An Accurate and Efficient Hybrid Mamba-Transformer Reasoning Model
- Llama-Nemotron: Efficient Reasoning Models
- Puzzle: Distillation-Based NAS for Inference-Optimized LLMs
- Self-Consistency Improves Chain of Thought Reasoning in Language Models
- Evaluating Large Language Models Trained on Code
- Large Language Monkeys: Scaling Inference Compute with Repeated Sampling
- Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters
- Resample or Reroute? Recoverable Stopping Debt Without Identified Action Selection
- The Curious Case of Neural Text Degeneration
- Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs
- Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models
- Measuring Mathematical Problem Solving With the MATH Dataset
- Let's Verify Step by Step
- Efficient Memory Management for Large Language Model Serving with PagedAttention
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis