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
arXiv:2606. 19162v1 Announce Type: new Abstract: Score- and flow-matching models often rely on preference-based reinforcement learning for two purposes: aligning with subjective preferences and, surprisingly, recovering properties such as visual realism and coherent object structure that matching-based training is intended to learn from the data itself.
By Nicolas Beltran-Velez, Felix Friedrich, Zhang Xiaofeng, Reyhane Askari-Hemmat, Xiaochuang Han, Adriana Romero-Soriano, Michal Drozdzal
arXiv:2607. 14408v1 Announce Type: new Abstract: A self-evolving agentic loop repeatedly proposes a tweaked version of an agent (its prompt template or program) and accepts or rejects the change based on a per-iteration quality signal.
By Minghao Liu, Yu Wang, Jiayun Wang, Wei Wei
arXiv:2606. 03980v1 Announce Type: new Abstract: Reward models (RMs) provide critical feedback signals for LLM post-training, notably in reinforced fine-tuning (RFT) and reinforcement learning (RL) pipelines.
By Tao Chen, Gangwei Jiang, Pengyu Cheng, Siyuan Huang, Yihao Liu, Jingwei Ni, Jiaqi Guo, Mengyu Zhou, Kai Tang, Junling Liu, Qinliang Su, Xiaoxi Jiang, Guanjun Jiang
The paper presents a post‑training approach for text‑to‑image models that combines a preference reward, trained on large human preference data, with rubric‑based rewards that assess prompt faithfulness and other desirable traits. The authors show that a simple reward composition strategy outperforms a naive weighted average, leading to significant Elo gains on the Arena leaderboard for models like Flux2dev and Ideogram‑4. They also release Arena‑T2I‑Training, a 1K subset of data to aid reproducible research in post‑training.
By Yuanhao Ban, I-Hung Hsu, Anastasios Angelopoulos, Wei-Lin Chiang, Ion Stoica, Cho-Jui Hsieh
SPADE (Self-Play in Adaptive Synthetic Executable Environments) is a reinforcement‑learning framework where a single large language model acts as both an Environment Designer—creating executable, long‑horizon training environments—and a Reasoning Agent—learning to act within those environments. The framework uses a regret signal based on the difference between rewarded performance with and without privileged hints to guide the Designer toward environments that are challenging yet solvable. Experiments show that, when scaled to 30‑billion‑parameter models, SPADE outperforms fixed‑environment baselines by significant margins across math, science, code, and reasoning benchmarks, and improves tool‑use performance on BFCL‑v4 and ACEBench‑Agent.
whyItMatters":"By making environment design a learnable component, SPADE enables continuous self‑improvement and demonstrates that adaptive, self‑generated training environments can substantially boost language‑model performance across diverse tasks."
By Bo Liu, Simon Yu, Yiding Jiang, Ao Qu, Andrew Zhao, Zichen Liu, Junsu Kim, Zijian Zhou, Seungone Kim, Tongzheng Ren, Mickel Liu, Hanfei Yu, Zhaorun Chen, Weiyan Shi, Paul Pu Liang, Luke Zettlemoyer, Yejin Choi, Natasha Jaques
The paper introduces a self‑evolving harness framework where a frozen language‑model agent first solves tasks and then edits its own harness based on run records. Using a 49‑line seed harness, the evolved harness improves average scores on in‑distribution benchmarks by 4.48 points and on out‑of‑distribution benchmarks by 12.64 points, surpassing Codex on the former and matching it on the latter. Continued evolution on a specific out‑of‑distribution benchmark further raises performance, and the study analyzes emergent mechanisms such as output truncation and history compaction.
By Qiankai Xu
Rubric-CEPR is a self‑evolving image editing framework that improves a pretrained editor using only its own generated samples, without human‑edited targets or external reward models. It employs a planner to generate structured edit instructions, an editor to produce multiple candidate edits, and a frozen critic that scores each candidate with rubric‑augmented contrastive edit‑preservation rewards, rejecting infeasible candidates via non‑compensatory gates. The best verified candidate is distilled into the editor through lightweight adapter training, yielding measurable performance gains on benchmarks such as Qwen‑Image‑Edit, GEdit‑Bench, and Complex‑Edit.
By Ritesh Thawkar, Shubham Patle, Shravan Venkatraman, Rao Muhammad Anwer
Reward models (RMs) provide critical feedback signals for LLM post-training, notably in reinforced fine-tuning (RFT) and reinforcement learning (RL) pipelines. However, current reward evaluation relies on heterogeneous criteria such as rule-based verifiers, ground-truth references, procedural checklists, and complex rubrics, where a unified mechanism to integrate all types of evidence remains unexplored.
arXiv:2512. 03438v3 Announce Type: replace Abstract: Agentic reasoning models trained with multimodal reinforcement learning (MMRL) have become increasingly capable, yet they are almost universally optimized using sparse, outcome-based rewards computed based on the final answers.
By Reuben Tan, Baolin Peng, Zhengyuan Yang, Hao Cheng, Oier Mees, Theodore Zhao, Andrea Tupini, Isar Meijer, Qianhui Wu, Yuncong Yang, Lars Liden, Yu Gu, Sheng Zhang, Xiaodong Liu, Lijuan Wang, Marc Pollefeys, Yong Jae Lee, Jianfeng Gao
arXiv:2604. 17415v3 Announce Type: replace-cross Abstract: Reward-based fine-tuning steers a pretrained diffusion or flow-based generative model toward higher-reward samples while remaining close to the pretrained model.
By Jeongjae Lee, Jinho Chang, Jeongsol Kim, Jong Chul Ye
VICO proposes a co‑evolutionary framework where a vision‑language model (actor) and an Environment‑as‑Rewriter (EnvRewriter) are trained together. The EnvRewriter edits visual structures—such as scene graphs, chart tables, or protected region masks—to generate new training samples whose difficulty is calibrated to the actor’s current performance via a pass‑rate reward. Across nine multimodal benchmarks, VICO‑8B outperforms its base model by up to +5.0 % on out‑of‑domain tasks and surpasses other co‑evolution baselines while requiring far fewer labeled samples.
By Meng Lu, Ligeng Zhu, Olivia Xiao, Yuchen Zhuang, Zihan Wang, Kuncheng Wu, Bangya Liu, Yu Wang, Charles Fleming, Wenqi Shi, Xuan Wang