SLCA-GRPO addresses cross‑segment credit misattribution in tool‑calling reinforcement learning by introducing Segment‑Locked Credit Assignment (SLCA), which separates advantage estimation for tool‑invocation tokens and natural‑language summary tokens. The method leverages a Schema‑Guided LLM Simulator (SGLS) for scalable training and Hierarchical Rewards (HierR) to route execution and preference advantages appropriately. Experiments on a 7B backbone show that SLCA‑GRPO outperforms baseline methods, improving in‑domain accuracy by 2.53 pp, the Berkeley Function‑Calling Leaderboard by 1.36 pp, and $ au^2$‑Bench by 9.15 pp while reducing tool redundancy and costs.
By Yan Zhan, Shaobo Liu, Qiunan Liu, Yuanjun Shi, Siqi Xu, WeiYi Hou, Xiang Xu, Zekang Li, Weizhou Pan, Jiahong Yan
arXiv:2609. 29140v1 Announce Type: new Abstract: Repeated evaluation can estimate a benchmark score accurately while still requiring replication to certify narrow uncertainty.
By Yezhou Cheng, Runjia Du, Zeming Liu, Qibai Chen, Hang Lyu, Yilan Wei, Yankai Zeng, Bojun Lin
The paper introduces PixelJev, a native-image decision interface that combines an image, a task instruction, and a runtime candidate set into a structured choice and candidate-conditioned probabilities using small open multimodal models. It unifies recognition and multiple-choice visual question answering via a language-model readout, offering options for frozen inference, language-side adaptation, and held-out calibration. Across seven benchmarks, 64-shot source adaptation significantly boosts Pets accuracy from 60.13% to 92.40%, and the system supports both VQA tasks with frozen inference, while also highlighting areas for improvement such as schema robustness and cross-family transfer.
By Xunlan Zhou, Xianliang Yang, Li Zhao
SWE-Prometheus is a new benchmark that evaluates large language model coding agents on the broader task of improving repository engineering governance, rather than just fixing individual issues. It presents fixed snapshots with open-ended objectives, requiring agents to identify risks, prioritize interventions, and verify changes across six governance dimensions. The benchmark includes 60 repositories and reports metrics such as Normalized Governance Improvement and behavior‑breakage rates, providing a nuanced view of how different models affect governance artifacts and execution‑backed improvements.
By Jiajun Wu, Leixin Sun, Zihan Tan, Yitao Liu, Shuo Li, Jiaru Qian, Shanghaoran Quan, Chuangxin Zhao, Yangxu Liao, Yang Liu, Bin Chong, Guancheng Wan
The paper introduces a method for safely retiring procedural guidance in AI agents that control physical actions. It proposes matched authority counterfactuals and a two‑gate retirement certificate to ensure that reductions preserve authorized utility while eliminating unauthorized protected effects. Experiments across multiple models and skill bundles show that task‑certified reductions can remove most skill clauses, but only a combined protocol passes both utility and safety gates in all tested configurations.
By Zhonghao Zhan, Xiao Ma, Hamed Haddadi
The paper introduces Jev, a reinforcement‑learning‑trained model that provides calibrated probability answers to typed questions about a single input in one call. Jev is evaluated on RLCDAlignBench, a benchmark covering ten alignment failures across 44 tests and five target models, achieving a median AUROC of 0.886 zero‑shot and outperforming supervised baselines on most tasks. The study shows that question wording has little impact, while contextual fields that encode labels are more influential, and that Jev matches human‑label agreement while being 63× cheaper than LLM‑judge scorers.
By Ruoqi Guo, Yi Liu, Gelei Deng, Yuekang Li, Lida Zhao, Yutao Wu, Simin Chen, Ying Zhang, Leo Yu Zhang
The paper investigates how sequential knowledge editing can degrade a language model’s ability to discern reliable evidence from unreliable evidence without affecting overall accuracy. Using a conservatively tuned LoRA on Qwen2.5‑7B‑Instruct, the authors show that after 1,000 edits the model’s arbitration score for untouched facts drops by 36%, leading to higher error rates on its most confident decisions, while MMLU accuracy remains unchanged. The study also finds that in some model‑method combinations, sequential edits can reduce MMLU to chance levels even though edit success and locality remain perfect.
By Atul Anand
The paper investigates whether reasoning always benefits universal multimodal embeddings (UMEs). By comparing the discriminative and reasoning-driven branches of UME-R1, the authors find that while reasoning improves positive similarity in 56.6% of cases, it also creates 15.7% false-helpful instances where hard negatives are drawn closer. Diagnostic analyses reveal that reasoning often de‑condenses retrieved neighborhoods and that chain‑of‑thought tokens encode evidence common to both positives and hard negatives. Based on these insights, the authors introduce SURE, a utility router that boosts UME-R1‑7B by 1.5 points and consistently improves other embedding models on MMEB‑V2 without retraining or extra VLM passes.
By Wenxiao Fan, Jingling Fu, Luohang Liu, Xinyuan Shan, Lichen Ma, Yu He, Junshi Huang, Yan Li, Kan Li
The paper introduces iCoder-27B, a 27‑billion‑parameter model for RTL design and GPU kernel optimization that is developed through a recursive AI‑led process with minimal human input. Human experts provide high‑level objectives and reusable research skills, while the agent autonomously selects experiments, diagnoses outcomes, and refines training strategies, coordinating SFT, self‑distillation, and reinforcement learning. iCoder outperforms GPT‑5.5 and Claude‑Opus‑4.8 on several benchmarks, demonstrating the feasibility of building frontier‑competitive models with largely automated development.
By Cheng Yang, Jiayang Lyu, Shangyuan Liu, Guibin Zhang, Jiong Lin, Xinlei Yu, Junchi Yan, Shuicheng Yan, Weinan E, Linfeng Zhang, Linfeng Zhang, Qibing Ren
EnSiTa is a trilingual multi‑domain parallel dataset and benchmark for English, Sinhala, and Tamil. It contains human post‑edited training data across seven domains and professionally translated test sets for those domains plus an additional one, all produced through a multi‑year, rigorously quality‑controlled process. The authors use EnSiTa to conduct a comprehensive study of domain‑specific machine translation across six language directions, comparing from‑scratch Transformers, pre‑trained models, and decoder‑only LLMs under various training‑data sizes, model scales, and domain settings.
By Surangika Ranathunga, Nisansa de Silva, Aloka Fernando, Kavindu Warnakulasuriya, Isuru Wijesiri, Menan Velayuthan, Charitha Rathnayaka, Thivaharan Varatharajan, Sajeevi Silva, Piumi Kandanaarachchi, Uthayasanker Thayasivam
Spot, Separate, and Enhance (SSE) is a multimodal, user‑guided generative model for audio remixing and enhancement. It rebalances audio, removes unwanted sources, and reduces reverberation using video and textual guidance. The authors introduce the DegradedMix dataset and adopt generative evaluation metrics, showing SSE outperforms existing baselines in controllability and remixing quality.
By Ilpo Viertola, Giulio Cengarle, Gouthaman KV, Daniel Arteaga, Lie Lu
Qwen‑Planner‑Agent is a closed‑loop AI‑for‑AI framework that enables large language models to act as both developers and participants in building advanced AI systems. The framework integrates data production, model training, and deployment through a shared action‑feedback‑verification contract, employing AI‑for‑Data, AI‑for‑Training, and AI‑driven model‑harness co‑evolution. It achieves top performance on MobilePA‑Bench by improving tool use, memory, skills, and sub‑agent coordination, while also showing gains on non‑mobile benchmarks.
By Tingyu Qu, Weigao Sun, Yuecheng Liu, Yucheng Zhao, Yi Zhu, Yifeng Ding, Qiyi Wang, Sihan Cao, Pengkun Jiao, Hanlei Xie, Xiongwei Wu, Qichao Wang, Haodong Zhang, Jiajun Liu, Yuhao Wang, Yuqing Xie, Junpeng Zhao, Long Chen, Ming Ma, Sihan Yang, Ziwang Zhao, Yanhao Jia, Liangquan Gong, Feida Zhu, Yiran Zhong, Steven Hoi
TTLab submitted a system for the AlexandriaX-2026 Subtask 3 on Arabic machine‑translation error‑span detection and classification. The approach treats the task as token‑level classification over surface forms, using focal loss with class weighting and dialect‑specific decoding thresholds to address label imbalance. MARBERTv2, among six Arabic pre‑trained encoders, achieved the best performance, ranking third overall, though classification of rare error types remains difficult, indicating a need for data augmentation.
By Ali Abusaleh, Bhuvanesh Verma, Alexander Mehler
HEXIS is a system that compiles agent skills into extended finite state machines, separating skill knowledge from control flow. It uses local instructions within states to guide reasoning and generation, while explicit transition conditions manage execution progress. The incremental compiler maps skill clauses and tool interfaces to state operations, aligns development traces to identify missing operations, and updates are validated through static checks and replay of traces, resulting in improved success rates and reduced execution tokens across benchmarks.
By Minghao LI
TopU-LBVS is a new multi‑target benchmark for ligand‑based virtual screening that addresses shortcomings of existing datasets by using hard‑negative decoys and a fixed 1:40 active‑to‑decoy ratio. It covers 93 protein targets across seven classes, provides three evaluation protocols (full, low‑data, and mini), and includes curated ChEMBL‑35 bioactivity data with property‑matched, structurally similar decoys. The benchmark demonstrates that performance drops sharply when moving from random‑decoy to hard‑negative evaluation, and it releases data, splits, code, and baseline implementations for reproducible comparison.
By Surbhi Kumar, Yuhe Zhou, Varun Shiralkar, Niu Huang, Baris Coskunuzer
This study evaluates automatic speech recognition (ASR) for adolescent health communication in Twi, Dagbani, and Ewe by benchmarking five ASR systems on a Bible corpus and a domain-specific ASRH dataset, then performing supervised domain adaptation with a fine‑tuned Qwen3-ASR-0.6B model. Fine‑tuning significantly lowered word and character error rates, especially for Ewe, and the adapted model was deployed in the KasaHealth voice‑first application, which received high user approval and highlighted remaining domain gaps. The work demonstrates that in‑domain data, rather than model size or computational resources, is the primary limitation for effective ASR in these languages.
By Stephen E. Moore, Akwasi Asare, Mich-Seth Owusu, Paul Azunre, Joel Budu, Lawrence A. Adu-Gyamfi
The paper introduces an acceptance protocol for sensor-coordinate and polarity binding in mechatronic commissioning, separating candidate generation from release authority. It employs a frozen four‑billion‑parameter local language model to handle requirements that a deterministic parser cannot support, and only releases plans when both facts can be derived by an external gate under a sealed grammar. Evaluation on 144 tasks showed that fabricated plans were correctly rejected, no false releases occurred among 83 releases, and incorrect user answers were identified in 169 of 431 pairings.
By Mehmet Iscan
RAPID is a system that automatically generates, verifies, and refines robot programs from a single visual human demonstration. It infers a testable task specification, action primitives, and an interactive environment, using an object-centric relational program representation to enable reuse beyond the demonstration. The approach was evaluated in simulation on eight contact-rich manipulation tasks and successfully deployed on a real Franka arm, showing strong generalization across object pose, shape, material, and environment.
By Yuyao Liu, Jiayuan Mao, David Hsu, Leslie Pack Kaelbling, Tom\'as Lozano-P\'erez
The paper introduces a three-level evaluation framework—behavioral deployment, LM-head readout, and probe recoverability—to distinguish whether a language model fails a syntactic test by not encoding structure or by failing to use it. Using a trilingual control-dependency benchmark, the authors find that probe recoverability consistently exceeds LM-head readout, which in turn exceeds behavioral deployment across seven models and three languages, with the largest gap observed in Qwen3-0.6B Instruct. Layer-localized activation patching shows that instruction tuning shifts the decoded layer later, suggesting decoding favors surface shortcuts and that behavioral evaluation understates what models encode while probing alone overstates what they deploy.
By Zhenyan Lu, He Wang, Xiaohui Huang
The paper introduces SVGLM, a framework that integrates scalable vector graphics (SVG) primitives into vision‑language models to enable image generation within reasoning tasks. By treating SVG both as image descriptions and text instructions, SVGLM offers a compact and interpretable method for connecting text and image reasoning. The authors provide a curated SVG‑based image editing dataset and demonstrate strong SVG generation and image‑aware reasoning performance on a mathematical benchmark.
By Sunli Chen, Ding Zhong, Ziqiao Ma, Jiaxin Liu, Zeyuan Yang, Hao Zhang, Lie Lu, Joyce Chai, Chuang Gan