FlyBy is introduced, a selective querying framework, and 4B and 8B variants are trained to reason first, diagnose what remains unresolved, and, at a knowledge bottleneck, query stronger models whose parametric knowledge extends beyond its own.
Abstract
Scaling test-time computation is a powerful way to improve language-model reasoning, and is particularly appealing for small reasoning models (sRMs) that are cheap to serve. However, is additional thinking always the right operation? By intervening at intermediate reasoning states across two model families and multiple scales, we find that self-refinement largely consolidates probability mass onto solutions already reachable from the current state, rather than making new ones reachable. These interventions reveal two failure regimes: execution bottlenecks, where the correct path is reachable and reflection can recover it, and knowledge bottlenecks, where relevant external information makes it reachable. Motivated by this distinction, we introduce FlyBy, a selective querying framework, and train 4B and 8B variants to reason first, diagnose what remains unresolved, and, at a knowledge bottleneck, query stronger models whose parametric knowledge extends beyond its own. Supervised fine-tuning bootstraps a multi-depth query action, and cost-aware reinforcement learning calibrates whether to query, what to ask, and how much to spend. On 1,158 hard problems across six benchmarks, FlyBy-4B achieves 45.96% pass@8, surpassing Qwen3-14B (41.64%) at 2.7 times lower serving cost, while also exceeding Qwen3-8B in pass@1 (16.85% vs. 15.31%). Scaling to FlyBy-8B further improves pass@8 to 51.81%.
This work investigates whether open-weight models can generate less verbose reasoning by changing the form of their own earlier thoughts, without retraining, and identifies reasoning history as one way to reduce future reasoning-token use and motivate model-specific compaction policies evaluated for both efficiency and...
A. Hristov· Proceedings of the Second Wo...· 0 citations
The results show that different thinking mechanisms can learn distinct computational solutions, even at similar ID performance, and in this setting, latent recurrence supports a reusable forward-search algorithm that generalizes beyond the training depth.
As the world changes, new information is published while older records remain online. This complicates reasoning for large language models (LLMs), which must retrieve evidence and resolve conflicts to determine which information remains valid. Existing evaluations treat information seeking as a black box, revealing lit...
Tian-Yi Hu, Bhavyajeet Singh, Niket Tandon et al.· 0 citations
Evaluated on two reasoning domains, math and multi-hop reasoning, across 1-8B models from different families, LoT consistently improves over KD and converges faster than staged curricula, highlighting the value of adaptive progression.
Ming-Hui Liu, Thomas Magelinski, De-Hao Yuan et al.· arXiv.org· 0 citations
Knowledge distillation transfers reasoning capabilities from large language models to lightweight students. However, distilled small models still struggle with multi-step reasoning. We observe that the distribution divergences between teacher and student models along reasoning paths are highly sparse, and target tokens...
Zhen-Xing Dou, Kun-Hua He, Ming-Yang Yi et al.· 0 citations
BELT (BELief-based Termination), which separates outcome prediction from stopping control from end-to-end controllers, and its accuracy-cost trade-offs are competitive with or better than those of end-to-end controllers.
Jingwei Ni, Dmitry Abulkhanov, V. Smirnov et al.· 0 citations