SPIRAL is a reinforcement‑learning framework that trains language models to employ three inference primitives—sequential reasoning within a trace, parallel sampling of independent traces, and aggregation of those traces—within a single compute pipeline. The model first generates multiple independent chain‑of‑thought traces in parallel, then produces a final aggregation trace conditioned on them, with all components optimized end‑to‑end for the reward of the aggregated response. Experiments on reasoning tasks demonstrate that SPIRAL scales efficiently with inference compute, achieving up to 11× better scaling efficiency and 15% higher performance compared to the GRPO baseline when all three primitives are scaled.
By Jubayer Ibn Hamid, Ifdita Hasan Orney, Michael Y. Li, Omar Shaikh, Yoonho Lee, Dorsa Sadigh, Chelsea Finn, Noah Goodman
arXiv:2512. 07843v2 Announce Type: replace-cross Abstract: Scaling inference-time computation has enabled Large Language Models (LLMs) to achieve strong reasoning performance, but their inherently sequential decoding incurs substantial latency, motivating parallelization of the generation process.
By Long Lian, Sida Wang, Felix Juefei-Xu, Tsu-Jui Fu, Xiuyu Li, Adam Yala, Trevor Darrell, Alane Suhr, Yuandong Tian, Xi Victoria Lin
The paper introduces POISE, a reinforcement learning algorithm that uses a model’s internal states as a value estimator to reduce variance in reinforcement learning with verifiable rewards (RLVR). By employing a lightweight probe that reads internal signals during the forward pass, POISE predicts baselines online and uses a cross‑rollout construction to keep gradients unbiased. Experiments on Qwen3‑4B and OLMo3‑7B‑Instruct‑DPO across six domains show POISE outperforms existing RLVR baselines, offering more stable training and a value model that generalizes across tasks and scales with the policy.
By Yunho Choi, Jongwon Lim, Woojin Ahn, Minjae Oh, Jeonghoon Shim, Yohan Jo
arXiv:2603. 25184v2 Announce Type: replace-cross Abstract: Reinforcement learning (RL) has become essential for post-training large language models (LLMs) in reasoning tasks.
By Jiahao Wu, Ning Lu, Shengcai Liu, Kun Wang, Yanting Yang, Bailong Lin, Chen Jason Zhang, Li Qing, Ke Tang
Achieving strong optimization generalization across diverse optimization problems while requiring limited training resources remains a challenging problem for optimization-oriented large language models (LLMs). Existing approaches typically rely on large-scale supervised datasets, costly reasoning annotations, and expensive intermediate step verification, resulting in substantial training overhead.
The paper introduces Decision-Flow Sampling (DF‑Sample), a training‑free, data‑free inference framework that builds a hierarchical reasoning tree, evaluates entire trajectories, and back‑propagates utilities to guide branching decisions. Unlike local step‑wise sampling, DF‑Sample explicitly assesses global paths, enabling it to recover high‑quality, low‑probability reasoning chains that standard decoding misses. On the GPQA benchmark, DF‑Sample attains 45.6% accuracy, outperforming power sampling (38.9%) and GRPO (39.9%) and consistently surpassing baselines across multiple models and benchmarks, demonstrating significant latent reasoning potential in pretrained LLMs.
By Zhendong Mi, Shaoyi Huang
ConsensusBench is a new dataset that supplies rule‑based process‑level signals for large language model reasoning. It identifies key intermediate conclusions—called Consensus Nodes—by filtering correct trajectories and clustering semantically equivalent statements. By incorporating a process reward derived from these nodes into GRPO‑style reinforcement learning, the authors create ConsensusPR, which reduces reward sparsity and improves performance on benchmarks such as AIME, GSM8K, and MATH‑500.
By Shi-Qi Yan, Chao-Hong Tan, Qian Chen, Wen Wang, Xiangang Li, Zhen-Hua Ling
arXiv:2606. 16222v1 Announce Type: new Abstract: Large Language Models (LLMs) increasingly rely on intermediate reasoning, yet explicit Chain-of-Thought (CoT) suffers from a linguistic space bottleneck: each thought must be decoded into tokens, causing high inference overhead.
By Xiandong Zou, Jing Huang, Jianshu Li, Pan Zhou
arXiv:2604. 23333v2 Announce Type: replace Abstract: Scaling test-time computation with reinforcement learning (RL) has emerged as a reliable path to improve large language models (LLM) reasoning ability.
By Liaoyaqi Wang, Chunsheng Zuo, William Jurayj, Benjamin Van Durme, Anqi Liu
The paper investigates whether the high costs of training chain-of-thought reasoning models can be reduced through algorithmic design. It introduces an autocurriculum approach that lets the model select which problems to focus on during training, showing that this method provably improves both supervised fine‑tuning and reinforcement learning. For supervised fine‑tuning, autocurriculum requires exponentially fewer reasoning demonstrations by targeting prompts where the model struggles, while for reinforcement learning it decouples computational cost from the quality of the reference model, making the burn‑in cost nearly independent of target accuracy.
By Nived Rajaraman, Audrey Huang, Miro Dudik, Robert Schapire, Dylan J. Foster, Akshay Krishnamurthy
arXiv:2606. 11209v1 Announce Type: cross Abstract: Visual question answering increasingly requires multi-step reasoning.
By Jingpei Wu, Xiao Han, Weixiang Shen, Boer Zhang, Zifeng Ding, Volker Tresp
GRIP (Granular Reward-guided Interpolation of Parameters) is a lightweight framework that blends a reasoning-oriented large language model with an instruction-tuned model by assigning learnable interpolation ratios to individual modules. The ratios are optimized while keeping both source models frozen, using a reward signal that prefers correct and concise responses. Experiments demonstrate that GRIP improves the accuracy-efficiency trade-off compared to fixed or search-based merging baselines and uncover module-wise fusion patterns linked to efficient reasoning.
By Lam So, Canhui Wu, Han Lin