The paper introduces Budget‑Efficient Thinking (BET), a two‑stage framework that treats adaptive reasoning as a computational investment, aligning solve‑or‑fold decisions with expected return rather than perceived difficulty. BET learns three distinct behaviors: concise short solves for easy queries, early abstention (nice fold) when further reasoning is unlikely to pay off, and allocating sufficient compute (hero call) for hard‑but‑solvable questions. Experiments on seven benchmarks with three base models show BET cuts reasoning tokens by 54% while boosting accuracy by up to 3.2%, and it transfers effectively to scientific QA and logical reasoning tasks.
By Zhaomeng Zhou, Lan Zhang, Junyang Wang, Mu Yuan, Songlin Liu, Tingzhao Li, Yiqing Hu, Yumeng Zhao
The paper evaluates how three large mixture‑of‑experts models (Alibaba, OpenAI, NVIDIA) can be fine‑tuned to reason in a low‑resource language, specifically Greek. Accuracy metrics show little change, but the authors uncover significant qualitative improvements: after supervised fine‑tuning, models reason in Greek on ~98% of items, with better grammaticality and retained general ability. Reinforcement learning with pre‑registered rewards further eliminates reasoning‑channel leaks and format skips, while the Greek‑reasoning habit remains robust to an accuracy‑only gradient.
By Ayoub Kirouane, Christos Petrocheilos
arXiv:2609.39111v1 Announce Type: new
Abstract: Human intelligence relies heavily on learned intuition: recognising patterns and judging situations without explicitly unfolding every intermediate ste...
By Li Ding, Haidi Jin, Chen Ji
Prefill‑only decision models evaluate every candidate in a menu in a single forward pass, avoiding decoding and reducing cost by one to two orders of magnitude compared to generative language models. The paper demonstrates that when only the candidate menu changes, the model’s post‑intervention accuracy can be predicted solely from the cached first‑pass distribution using a simple estimator that renormalizes and selects the argmax, without any labels or second pass. Across seven model families, ten datasets, and two task types, this menu‑only intervention prediction is within 4.2 points of actual accuracy, and in one family it is exact, whereas a probability‑level variant fails by 21 points, indicating the property resides in ranking rather than calibrated probabilities.
By Ran Li, Lei Chen
RecurTrace introduces adaptive latent reasoning for language models by allowing each looped layer to attend to its own past states and by using a halting head to decide when to stop iterating. This approach overcomes two limitations of prior latent recurrence methods: limited access to earlier computations and a fixed loop count that mismatches input difficulty. In experiments on MathQA, RecurTrace achieves 56.9% accuracy with an average of 2.0 loops, outperforming fixed‑depth baselines and other adaptive methods, and it also improves generation accuracy across a range of model sizes.
By Yuxiang Wang, Kunyu Feng, Yingda Shen, Haoning Xu, Junyu Wang, Zhizheng Wu
arXiv:2608. 15445v1 Announce Type: new Abstract: When a reward is correct on every training example yet consistent with more than one goal, a model can acquire an unintended one, a failure known as goal misgeneralization.
By Suyash Maniyar, Armaan Sandhu, Abhishek Mishra
arXiv:2602. 14265v3 Announce Type: replace-cross Abstract: Inference-Time-Compute (ITC) methods like Best-of-$n$ and Tree-of-Thoughts are meant to produce output candidates that are both high-quality and diverse, but their use of high-temperature sampling often fails to achieve meaningful output diversity.
By Zachary Bamberger, Till R. Saenger, Gilad Morad, Ofra Amir, Brandon M. Stewart, Amir Feder
arXiv:2607. 18100v1 Announce Type: new Abstract: Extended reasoning has become standard for frontier Large Language Models (LLMs), yet the trajectories these models produce remain largely uncontrollable.
By Sheldon Yu, Tong Yu, Xunyi Jiang, Rohan Surana, Gagan Mundada, Sungchul Kim, Lina Yao, Julian McAuley, Junda Wu
arXiv:2607. 14552v1 Announce Type: cross Abstract: A standard recipe for distilling the reasoning ability of large language models (LLMs) is to sample chains of thought from the model, keep those that reach the correct final answer, and fine-tune on the survivors.
By Jungseob Lee, Seungyoon Lee, Suhyune Son, Dongyub Jude Lee, Sungbin Han, Sugyeong Eo, Heuiseok Lim
arXiv:2609.16055v1 Announce Type: cross
Abstract: Test-time compute has emerged as a major approach to improving the capabilities of Large Language Models (LLMs). However, existing test-time reasonin...
By Zhiren Gong, Yikun Hou, Zihao Zeng, Ming Xiao, Chau Yuen, Wei Yang Bryan Lim
arXiv:2609.37119v1 Announce Type: cross
Abstract: Recent approaches to reinforcement learning (RL) post-training for large language models increasingly remove the critic to reduce training instabilit...
By Hongyang Li, Xiao Li, Caesar Wu, Said Mammar, Gr\'egoire Danoy, Pascal Bouvry
The paper introduces Drift-Constrained Optimization (DCO), a framework that treats behavioral drift during fine‑tuning of instruction models as a bounded constraint rather than an uncontrolled side effect. By defining a drift budget, the authors reformulate fine‑tuning as a direction‑selection problem, showing that choosing different update directions can qualitatively change outcomes. Experiments on Qwen3 models demonstrate that carefully selected directions improve scientific reasoning and multilingual translation while preserving reasoning capabilities and general performance.
By Fei Yuan, Changjiang Gao, Yilei Tu, Yifeng Liu, Shujian Huang, Yu Qiao