The paper introduces a visual-native agent harness that registers every tool‑returned image as an addressable reference, enabling reuse of intermediate visual evidence in multimodal deep search tasks. It also proposes On‑Policy Data Evolution (ODE), a closed‑loop data generator that refines training data across rounds based on policy rollouts, targeting the agent’s evolving needs. Experiments on eight benchmarks show ODE boosts the Qwen3‑VL‑8B agent’s performance from 24.9% to 39.0% on average, and from 30.6% to 41.5% at 30B, while validating the benefits of image bank reuse and rollout‑feedback evolution.
By Shijue Huang, Hangyu Guo, Guanting Dong, Chenxin Li, Junting Lu, Xinyu Geng, Zhaochen Su, Zhenyu Li, Shuang Chen, Hongru Wang, Yi R. Fung
arXiv:2608. 03979v1 Announce Type: cross Abstract: We introduce Video-DeepResearch (Video-DR), extending multimodal agents from static images to continuous video streams, a setting that demands dense spatiotemporal grounding coupled with open-web exploration.
By Zhen Fang, Yu Zeng, Wenxuan Huang, Yiming Zhao, Shiting Huang, Tianfei Ren, Qi Lu, Qingnan Ren, Qisheng Su, Lionel Z. Wang, Qingyu Yin, Shuang Chen, Zehui Chen, Lin Chen, Zhenfei Yin, Yao Hu, Shaohui Lin, Wanli Ouyang, Shaosheng Cao, Feng Zhao
The paper presents an unsupervised multimodal reward model built on a frozen Qwen3‑VL backbone. By adding a linear head to an intermediate layer and training it to rank a web caption higher than a perturbed version of itself, the authors avoid human preference labels and achieve competitive performance—outperforming GPT‑4o on several benchmarks with far fewer parameters and compute. The approach demonstrates that a carefully chosen intermediate layer, simple perturbation-based training, and verifier‑driven fine‑tuning can yield effective vision‑language rewards without costly annotation.
By Yueying Li, Jingxuan Fan, Kiant\'e Brantley, Sham M. Kakade, Hanlin Zhang
The paper introduces Reinforcement Learning with Verifiable Rewards (RLVR) applied to small search agents, specifically training a Qwen3.5-0.8B model with Group Relative Policy Optimization and an interleaved Wikipedia-search tool on the MuSiQue dataset. Experiments varying reward shapes across three seeds show that RLVR can achieve a 3.8‑fold improvement over an untrained baseline, with the best run reaching a 0.352 average exact match. The study finds that the sparse exact‑match reward, standard in larger models, performs poorly for small models, indicating that reward design must be tailored rather than scaled down from large‑model recipes.
By Gaurisankar Jayadas, Aske Plaat, \'Alvaro Serra-G\'omez, Sandheep P
arXiv:2608.30705v1 Announce Type: new
Abstract: Multimodal large language models (MLLMs) struggle with fine-grained Visual Search, the task of locating small or rare objects in high-resolution images...
By Jingyi He, Sanghwan Kim, Zeynep Akata
arXiv:2606. 05597v3 Announce Type: replace Abstract: Training vision-language web agents with multi-step RL is compute-intensive, with two dominant forms of inefficiency: idle GPUs in synchronous RL, and trajectories that use more steps and tokens than necessary.
By Hao Bai, Rui Yang, Chenlu Ye, Spencer Whitehead, Aviral Kumar, Tong Zhang
We present SimpleSearch-VL, an efficient, reliable, and practical framework for multimodal agentic search. Its core idea is to improve the agent's own search-and-verification process rather than scaling data, tools, or auxiliary model components.
arXiv:2608. 00335v1 Announce Type: cross Abstract: Compact web agents can reduce deployment cost, but training them poses challenges in both data collection and post-SFT reinforcement learning (RL).
By Chengbo Liu, Lifang Zhou, Ruijie Yan, Pei Tan, Ao Sun, Haojun Huang, Guichun Hua, Sining Wei, Yining Chen, Yingying He, Yutao Xie
arXiv:2608. 01205v1 Announce Type: new Abstract: Recent offline reinforcement learning methods increasingly rely on expressive generative policies and specialized value-guidance mechanisms.
By Denis Tarasov, Robert K. Katzschmann
The paper introduces WebMRE, an offline benchmark comprising 541 tasks and 5,293 steps extracted from WebArena trajectories, designed to provide deterministic scoring for web agents without live environments. It enables the first systematic study of how guide sentences and grounded actions reinforce each other, showing that jointly decoding a guide improves element selection accuracy and that the guide acts as a causal instruction channel. The authors fine‑tune models that outperform leading zero‑shot baselines on all offline metrics.
By Chengguang Gan, Yunhao Liang, QingHao Zhang, Shiwen Ni
The paper investigates how trajectory fine‑tuning can enhance small language models (SLMs) as next‑action controllers in retrieval‑augmented question answering. By building a seven‑way action‑prediction task from teacher search traces, the authors fine‑tune SLMs and cross‑lingual SLMs (xSLMs) using LoRA and evaluate on 1,646 held‑out examples, achieving a macro‑F1 of 0.6536 with Granite 4.1 3B. In an end‑to‑end controller/generator swap experiment on 149 trajectories, the fine‑tuned model improves Exact Match from 0.7530 to 0.7946 and token F1 from 0.7783 to 0.8295, demonstrating that trajectory supervision boosts action prediction and evidence‑recording behavior.
By Mohammed Al-Maamari, Saber Zerhoudi, Michael Granitzer, Jelena Mitrovi\'c
The paper introduces VHOP, a data generation framework and benchmark for visual agentic search, and VHOP-Router, an end‑to‑end training pipeline that turns a standard embedding model into an autoregressive multi‑step retriever operating directly in visual latent space. VHOP-Router eliminates the need for intermediate text queries, boosting retrieval accuracy from under 5% to 76.3% and improving task success rates by 52.7% while dramatically reducing token usage and API payloads. The approach generalizes to unseen difficulty levels and realistic test sets, offering an efficient solution that preserves native LLM capabilities.
By Tianyu Chen, Mingyuan Zhou, Jiaxing Wu