DAGent introduces an Evaluate‑then‑Grow planning approach for deep research agents, building directed acyclic graphs incrementally based on confidence and uncertainty from completed tasks. The framework includes a hierarchical context layer for efficient query handling and a structural reinforcement learning component, DAGRPO, that rewards topology‑conditioned execution. Experiments on BrowseComp‑Plus, GAIA, and xbench‑DeepSearch show DAGent outperforming strong baselines across multiple backbones and scaling to large language models.
By Hanwen Liu, Yuanfu Sun, Qiaoyu Tan
The paper introduces Mosaic, a multi-image visual harness that lets large language‑vision models (MLLMs) construct visual intermediates using ten composable image operations. It evaluates five re‑representation settings on existing multi‑image benchmarks and a new grounding‑focused benchmark, MosaicBench, finding that visual re‑representation benefits tasks requiring precise visual evidence more than those dominated by high‑level semantics. MosaicAgent‑8B is trained via reinforcement learning to compose these operations without demonstration trajectories, demonstrating diverse problem‑solving patterns.
By Gengyuan Zhang, Xiao Han, Xinyu Xie, Tong Liu, Volker Tresp
The paper introduces On‑Policy Warmup (OPW), a teacher‑guided training stage where a student agent learns from a teacher on its own interaction trajectories before switching to reinforcement learning with verifiable rewards (RLVR). OPW differs from traditional imitation by focusing on states generated by the student’s own decisions, including imperfect actions and recovery situations. The authors provide a theoretical link between on‑policy reverse‑KL distillation and trajectory‑level distribution matching, showing that, under a competent teacher and low distillation loss, OPW can lower bound initial verifier success and reduce reward‑discovery complexity, thereby accelerating RLVR performance.
By Yitong Qiao, Tiantian He, Lei Liu, Yue Shen, Jian Wang, Jinjie Gu, Zhixuan Chu
CAST introduces a reinforcement‑learning fine‑tuning framework for diffusion models that addresses three key limitations: it automatically selects the denoising window based on each model’s trajectory, decomposes prompts into verifiable semantic atoms via Causal Scene Graphs, and applies atom‑level rewards spatially weighted in the policy objective. The method is applied to FLUX.2‑dev and Qwen‑Image‑2512, yielding up to 3.07× improvement on the hardest GenEval 2 prompts compared with Flow‑GRPO while also enhancing overall generation quality.
By Shu Yu, Chaochao Lu
OmniReasoning introduces a new benchmark, OmniReasoningBench, that requires both audio and visual evidence for answering 1,150 multiple-choice and open-ended questions across two tasks. The authors also develop OmniQA, a data engine that automatically generates evidence‑grounded QA pairs with time‑stamped clue chains, producing training datasets OmniReasoning‑SFT‑112K and OmniReasoning‑RL‑19K. Finally, they propose Modality‑Factored Self‑Distillation (MFSD), an on‑policy self‑distillation method that assigns token‑level credit by evaluating responses under modality‑specific clue contexts, enabling the OmniReasoning‑30B‑A3B model to achieve significant performance gains on both the new benchmark and existing video benchmarks.
By Junming Lin, Yuxuan Wang, Zhenxin Lei, Yuxin Liu, Ruixun Liu, Yinsong Yan, Ling Wang, Minghao Han, Yunfei Chu, Shun Lei, Xueyao Zhang, Qize Yang, Jin Xu, Yiwu Zhong
GroundingPI is a 4‑billion‑parameter grounding foundation model that generates points and boxes as quantized coordinates in a shared vocabulary. It is trained with multimodal and spatial pretraining, supervised fine‑tuning, and reinforcement learning, achieving a new state‑of‑the‑art average of 73.68% across 34 grounding benchmarks. As a visual backbone, GroundingPI improves performance in robotic manipulation and autonomous driving, outperforming larger models and mainstream backbones in several out‑of‑distribution settings.
By Qize Yu, Lianrui Fan, Boyu Chen, Jiaqi Liang, Xini Ding, Yue Chen, Zetian Song, Yuran Wang, Yi Zou, Kaixuan Wang, Tianxing Chen, Wenxuan Song, Bohan Zhou, Mingleyang Li, Siqiao Huang, Yuqi Ye, Caigao Jiang, Wei Wei, Ruihai Wu, Hang Zhang, Yixiao Ge, Shuchang Zhou, Shilong Liu, Xianming Liu, Ping Luo, Shiyu Huang
The paper investigates whether enhancing goal representations improves goal-conditioned reinforcement learning (GCRL) performance. By creating an exact temporal-distance goal representation in deterministic mazes and systematically degrading its geometric quality, the authors find that changes in goal representation have little effect on performance. In contrast, degrading the agent’s current state representation more than doubles failure rates, indicating that state representation is the critical bottleneck. The study further demonstrates that simple random Fourier positional encodings can significantly boost performance on challenging navigation tasks without additional map or objective modifications.
By Syed Nazmus Sakib, Abdul Monaf Chowdhury, Nafiul Haque, Shifat E Arman, Md Mehedi Hasan
PhantomEnvironments is a framework that trains large language model agents in synthetic, rule‑generated fictional worlds. By creating multi‑turn reinforcement learning environments where agents search templated articles to answer multi‑hop questions, the approach eliminates the need for costly human data or hallucinated LLM‑generated settings. Agents trained in these zero‑cost, purely rule‑based worlds transfer effectively to real‑world multi‑hop search benchmarks, often surpassing models trained on real data, and demonstrate scalable search behavior that grows linearly with question difficulty.
By Anmol Kabra, Swathi Saravana Selvam, Albert Gong, Chao Wan, Christian Belardi, Dongyoung Go, Katie Z. Luo, Kilian Q. Weinberger
arXiv:2609.40360v1 Announce Type: cross
Abstract: Reinforcement learning with verifiable rewards (RLVR) has improved the reasoning capabilities of large language models (LLMs), yet their predictions...
By Junshu Pan, Zhizhang Fu, Shulin Huang, Yiran Ding, Zifan Cheng, Wenqi Shao, Qiaosheng Zhang, Yue Zhang
arXiv:2609.00768v2 Announce Type: replace
Abstract: Self-play supports the self-evolution of language models, but solver performance can plateau or decline across rounds without guidance. Existing un...
By Xincheng Wei, Yifan Ding, Fucheng Xiong, Yoshua Li, Dongsheng Ma, Rongxiang Weng, Xunliang Cai, Wenjian Ding, Yao Zhang
arXiv:2609.33665v2 Announce Type: replace
Abstract: Automatically generated environments provide a scalable source of interaction data for training general agents. However, existing approaches mainly...
By Xiao-Wen Yang, Weiyi Xu, Wen Da, Hang Xu, Canwei Li, Hong-Jie You, Pusen Dong, Yucheng Zeng, Zhaokai Luo, Yu-Feng Li, Yao Hu, Mu Chuan
arXiv:2601.18681v3 Announce Type: replace-cross
Abstract: We consider time discretization for score-based diffusion models to generate samples from a learned reverse-time dynamic on a finite grid. Un...
By Yilie Huang, Wenpin Tang, Xunyu Zhou
arXiv:2604.09107v2 Announce Type: replace-cross
Abstract: Modern LLM reinforcement learning (RL) workloads require a high-performance weight transfer system to scale training across heterogeneous com...
By Chenhao Ye, Huaizheng Zhang, Mingcong Han, Baoquan Zhong, Xiang Li, Qixiang Chen, Xinyi Zhang, Weidong Zhang, Kaihua Jiang, Wang Zhang, He Sun, Wencong Xiao, Andrea C. Arpaci-Dusseau, Remzi H. Arpaci-Dusseau
arXiv:2605.29307v2 Announce Type: replace-cross
Abstract: Large Language Model (LLM) search agents have shown strong promise on knowledge-intensive tasks through iterative reasoning and retrieval. Mo...
By Alireza Salemi, Chang Zeng, Atharva Nijasure, Jui-Hui Chung, Razieh Rahimi, Fernando Diaz, Hamed Zamani
arXiv:2607.02137v3 Announce Type: replace-cross
Abstract: We study timestep allocation for score-based diffusion sampling, where a learned reverse-time dynamics is discretized on a finite grid. Unifo...
By Yilie Huang, Wenpin Tang, Xun Yu Zhou
arXiv:2609.34422v2 Announce Type: replace-cross
Abstract: Language-model agents increasingly tackle long-horizon tasks whose interaction histories exceed the model's active context. Recent work has b...
By Lirui Luo, Kelong Mao, Heming Xia, Rongqing Li, Xinwei Yang, Luyu Chen, Kieran Wong, Yudong Guo, Xinrui Wang, Jiayin Zhu, Simiu Gu, Sulong Xu, Cong Fang
arXiv:2609.38329v1 Announce Type: new
Abstract: Group-relative RL methods such as Flow-GRPO post-train image generators by exploring with isotropic Gaussian noise added at every denoising step. This...
By Shuyue Stella Li, Xiaochuang Han, Yulia Tsvetkov, Luke Zettlemoyer
arXiv:2609.38428v1 Announce Type: new
Abstract: Real-time commentary for Multiplayer Online Battle Arena (MOBA) esports requires a vision-language model (VLM) to narrate a live match second by second...
By Shengyun Zhong, Xinkang Zhao, Ziyuan Chu, Linchao Zhu
arXiv:2609.38641v1 Announce Type: new
Abstract: Vision-Language-Action (VLA) foundation models have recently emerged as one of the prevailing solutions for autonomous driving, as they can utilize kno...
By Kai Yan, Xiangyu Chen, Yulong Cao, Alex Naumann, Peter Karkus, Yan Wang, Jef Packer, Alex Schwing, Yuxiong Wang, Boris Ivanovic, Wenjie Luo, Marco Pavone
The paper introduces Stackelberg Alignment, a leader‑follower framework that lets a pool of language models collaborate and improve by learning from each other’s responses. An EXP3 bandit leader adaptively selects instructions based on difficulty and discriminability, while the models act as followers, evaluating peers and learning via DPO or GRPO with Elo‑style reputation weighting and opponent matching. Experiments on diverse benchmarks show that this adaptive curriculum outperforms static baselines by up to 12‑25% and improves multi‑LLM evolution.
By Christina Hahn, Shangbin Feng, Dean Light, Swastik Roy, Hila Gonen, Yulia Tsvetkov