The paper presents a method for turning expert diagnoses of verification failures into reusable guidance for coding agents. By combining executable language definitions in the K framework with a set of procedures for constructing specifications, repairing proofs, and auditing their adequacy, the authors achieve a 164/164 success rate on the HumanEval benchmark after two targeted repairs. They further demonstrate that audits can detect defects missed by successful proofs and evaluate the approach on KleverBench and Optimism proofs, highlighting both progress and remaining challenges.
By Yuqing Zhai, Xiaohong Chen, Lingming Zhang, Sriram Vishwanath, Grigore Rosu
ComputerSD is an online self‑distillation method for computer‑use agents that leverages real‑time feedback from executed GUI transitions. It uses a fine‑tuned GUI analyzer to generate guidance and a step‑level value score after each action, combining token‑level OPSD with trajectory‑level GRPO in an asynchronous training framework. On the OSWorld‑Verified benchmark, ComputerSD improves performance over outcome‑only GRPO by 1.9 and 4.1 percentage points on Qwen3‑VL‑8B‑Thinking and EvoCUA‑8B backbones, and shows strong generalizability in out‑of‑distribution tests.
By Yong Du, Tongbo Chen, Zhengxi Lu, Yizhou Liu, Bofan Chen, Tao Jiang, Wenhao Xu, Yongliang Shen
arXiv:2609.38383v1 Announce Type: cross
Abstract: Random exploration reveals how an environment can be traversed before a goal is specified. Can this experience support long-range planning without po...
By Deqian Kong, Guangyan Sun, Sheng Cheng, Sirui Xie, Bo Pang, Jianwen Xie, Tony Geng, Caiwen Ding, Ying Nian Wu
arXiv:2609.38536v1 Announce Type: cross
Abstract: Diffusion-based large language models (dLLMs) promise to break the sequential latency bottleneck of autoregressive agents through parallel decoding,...
By Jiacheng Qiu, Christopher E. Mower, Jan Peters, Haitham Bou-Ammar, Matthieu Zimmer
The paper introduces the concept of persistent calibration, which requires a confidence estimator to accurately reflect a model’s evolving knowledge without additional supervision. It evaluates this by comparing confidence estimators trained on earlier checkpoints to their performance on later checkpoints using knowledge contrast sets—questions that shift from correct to incorrect answers across checkpoints. The study finds that standard inference-time and fine-tuning methods underperform compared to oracle methods, and suggests that multi-checkpoint training can improve calibration by identifying robust confidence features.
By Victor Wang, Thomas Hofweber, Mohit Bansal, Elias Stengel-Eskin
The paper introduces Length Self-Distillation (LSD) to address the length‑scaling tax (LST) that occurs during reinforcement‑learning post‑training, where models produce unnecessarily verbose responses to already‑solved prompts. LSD routes solved prompts to an on‑policy distillation process while keeping the original RL objective for unsolved prompts, using an exponential moving average of the online policy as its teacher. Experiments show LSD matches or surpasses RL performance while reducing LST from 19.0% to –3.7% on single‑turn reasoning and from 31.4% to 13.7% on multi‑turn agentic tasks, thereby maintaining concise responses on easy queries while still enabling exploration on difficult ones.
By Xu Wan, Wenyue Xu, Shengjie Zhao, Mingyang Sun
The paper introduces OptTips, a knowledge base of 50 expert optimization modeling techniques, and OptDachshund, a multi‑agent framework that generates mathematical models and solver code from natural‑language problem descriptions. Using these tools, the authors create the EfficientOpt benchmark, comprising 561 expert‑reviewed tasks with paired reference implementations, to evaluate large language models (LLMs) on both correctness and computational efficiency. Their experiments with 11 LLMs show a consistent efficiency gap: even when LLMs produce correct solutions, the resulting programs often take longer to solve than expert‑crafted counterparts, highlighting the need to assess both accuracy and runtime performance in LLM‑based optimization modeling.
By Zhong Li, Xin Huang, Jinhui Wan, Xiangyi Wang, Shenkai Zhang, Ruiqi Chen, Wenyu Liu, Zaiwen Wen, Ziyan Luo
The paper proposes an inference auction for large language model (LLM) APIs, enabling users to bid for faster service when compute demand exceeds capacity. The auction allocates priority efficiently without increasing latency, and includes fast algorithms for truthful bidding and an autobidding agent that adjusts bids within a user’s budget to maximize utility. Experiments show the auction improves system welfare while preserving the cache utilization and latency benefits of the SGLang inference framework.
By Keegan Harris, Siddharth Prasad, Asher Trockman, Nika Haghtalab, Michael I. Jordan
arXiv:2609.40137v1 Announce Type: cross
Abstract: We present Game-Guided Skill Discovery (GGSD), a framework that uses self-play in games to discover motor skills that are directly playable by humans...
By Seungeun Rho, Jeonghwan Kim, Xue Bin Peng, Sehoon Ha
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
arXiv:2609.38604v1 Announce Type: cross
Abstract: Modern LLM agents increasingly tackle complex tasks through interactive, long-horizon exchanges with users, while existing benchmarks generally assum...
By Zheyuan Zhang, Mengyuan Chao, Ke Xiao, Ziyi Chen, Daoan Zhang, Yan Zhang, Yanfang Ye, Wei Xu
arXiv:2609.38653v1 Announce Type: cross
Abstract: Recent advances in musculoskeletal modeling and reinforcement learning have enabled muscle-actuated agents to reproduce increasingly complex human mo...
By Merkourios Simos, Chengkun Li, Bianca Ziliotto, Alexander Mathis
arXiv:2609.38914v1 Announce Type: new
Abstract: Evaluating interactive agents is expensive. Agent behavior is stochastic, so reliability must be measured over repeated trials, but failures are rare a...
By Priyanath Maji, Spandan Ghose Chowdhury
arXiv:2609.38982v1 Announce Type: cross
Abstract: Coding agents powered by large language models (LLMs) have shown remarkable abilities to autonomously reason about and achieve goals in the digital w...
By Jiaheng Hu, Roberto Martin-Martin, Peter Stone, Rocky Duan, Zhenyu Jiang, Guanya Shi
The paper introduces FAME, a training‑free framework that evaluates false memory in autonomous agents by tracking how their internal beliefs shift under counterfactual scenarios. False memory, defined as biases arising from spurious correlations, environment shifts, or knowledge conflicts, is hard to detect with standard methods. FAME measures concept drift in hidden states, achieving AUROCs between 76.2% and 96.7% and outperforming baselines on benchmarks such as GSM‑Symbolic, GitChameleon, and BigBench‑Hard.
By Quan M. Tran, Zhuo Huang, Zhen Fang, Jing Zhang, Mingming Gong, Tongliang Liu
arXiv:2609.39645v1 Announce Type: cross
Abstract: Multi-agent collaboration lets large language models (LLMs) improve question answering through deliberation and feedback. Yet shared discussion coupl...
By Weijie Ren, Yanwen Zhang, Hao Li, Zhuolin Qi, Hengyi Zhang, Naibo Wang
arXiv:2604.16778v3 Announce Type: replace-cross
Abstract: Modern agents specialize in varying domains while there is no clear approach combining different domain skills. We propose a federated learni...
By Dixi Yao, Tahseen Rabbani, Manzil Zaheer, Tian Li
arXiv:2511.10661v2 Announce Type: replace
Abstract: It is increasingly important to evaluate the characteristics of systems based on large language models (LLMs). Evaluations in this context often re...
By Saatvik Kher, Shang Wu, Rachel Longjohn, Catarina Bel\'em, Padhraic Smyth
arXiv:2603.01209v3 Announce Type: replace
Abstract: In CodeAct, language-model agents write Python that calls tools and use execution feedback to choose actions. Persistent runtimes preserve Python v...
By Victor May, Van Khue Nguyen, Aaditya Salgarkar, Yishan Wang, Diganta Misra, Huu Nguyen
arXiv:2603.28013v4 Announce Type: replace-cross
Abstract: Multi-agent LLM systems now read documents, web pages and tool results on behalf of users, yet their resistance to prompt injection is usuall...
By Haochuan Kevin Wang, Zechen Zhang