The paper introduces a coroutine-bridge harness that lets a language model emit a Python program to manage tool calls in the CAR-bench evaluation. By decoupling model invocations from tool round-trips, the approach reduces model calls to a median of two per task while maintaining seven agent turns, achieving a median latency of 1.8 s on a Cerebras gpt‑oss‑120b. The harness achieved 60.0 % Pass³ on the official hidden evaluation, outperforming the baseline by 4.5× and matching frontier-model agents on GPT‑5.5, all while keeping the prompt largely cached and minimizing input compute.
By Ivan Matveev
arXiv:2608. 15089v1 Announce Type: new Abstract: Long-horizon agents can fail even when their underlying models can solve the constituent steps.
By Ziheng Qin, Yaxin Lu, Zhangyang Atlas Wang, Kai Wang
arXiv:2608.22510v1 Announce Type: new
Abstract: Agent benchmarks often evaluate only final answers even when agents run on stateful runtimes. We argue this under-specifies what is being evaluated: th...
By YuanHang Xiao
The paper introduces NP‑Bench, a benchmark and a proactive scheduling planner that coordinates parallel large‑language‑model coding agents. By partitioning work scopes and ordering merges ahead of time, the planner improves clean‑integration rates from 1/9 to 9/9 and eliminates merge conflicts, outperforming both no‑coordination and reactive‑detection baselines. It also demonstrates that cross‑session memory can eliminate repeated mistakes and that routing facts to agents does not improve long‑context accuracy at scale.
By Sumanyu Muku
The paper discusses the problem of benchmark maintenance debt in evaluating coding agents, where benchmarks become outdated as models improve and infrastructure changes. It introduces a continuous‑validation workflow that incorporates community feedback, automated checks, and human repairs, applied to Terminal‑Bench 2.0. The audit uncovered issues in about 30% of the benchmark, leading to a new version (Terminal‑Bench 2.1) that improved accuracy by over 10 percentage points and altered leaderboard positions, while also highlighting the importance of validating resource budgets alongside task artifacts.
By E. Kelly Buchanan, Crystal Zhou, Scott W. Linderman, Christopher R\'e
InfraDev is a new benchmark comprising 36 tasks derived from real-world pull requests to Megatron‑LM, vLLM, and verl, designed to test coding agents on infrastructure development for training, inference, and reinforcement learning across many GPUs. Each task requires agents to reimplement multi‑file changes and run correctly, with grading that includes hidden probes and a prompt exposing only the interfaces used. The benchmark reveals that current frontier models resolve only about 39% of tasks, with probes rejecting a third of submissions that pass official tests, and that structural and state errors dominate failures.
By Yangshen Deng, Jiaqi Ruan, Yeqi Huang, Nick Lau, Guangming Sheng, Luo Mai
The paper introduces CARE-PO, a method for controlling tool‑call efficiency in large language model agents. By anchoring an efficiency reward to a fixed reference budget tied to task difficulty and injecting successful trajectories into training rollouts, CARE‑PO prevents the collapse to zero tool calls and allows a single training run to produce multiple operating points balancing efficiency and capability. Experiments on the GAIA benchmark and other multi‑hop tasks show that CARE‑PO improves both success rates and call counts across several model sizes.
By Xiaoyang Ye, Xiao Yao, Hairui Zhu, Yiying Yang, Bei Gan, Jinliang Lu, Tengjin Weng, Yuan Yao, Wenhao Jiang
arXiv:2610.14146v1 Announce Type: new
Abstract: Long-horizon terminal benchmarks measure whether an agent can sustain progress through extended workflows with process rewards, not only whether its fi...
By Zongxia Li, Zhongzhi Li, Yucheng Shi, Ruhan Wang, Junyao Yang, Chengsong Huang, Zhichao Liu, Xiyang Wu, Anhao Li, Yue Yu, Ergan Shang, Fuxiao Liu, Tong Zheng, Ninghao Liu, Lichao Sun, Haitao Mi, Leowei Liang
arXiv:2606. 17929v1 Announce Type: new Abstract: Computer-using agents drive real software through the screen -- clicking and typing -- but they solve every task from scratch: asked to repeat a task, an agent re-reads the screen, re-reasons every tap, and pays the full cost again.
By Bojie Li
Software engineering (SWE) agents resolve repository-level issues through long trajectories that grow increasingly expensive as context accumulates. Failed runs tend to be longer and exhibit redundant exploration or looping, suggesting that some failures may be detectable before completion.
UniACE is a unified framework that standardizes the evaluation of large language model (LLM) agents by representing each benchmark as an instruction–tool–environment triplet and running models through a shared, task‑agnostic harness in isolated runtimes. It preserves native success criteria, offers an offline mode for dynamic‑resource tasks, and standardizes efficiency metrics, execution records, and failure attribution. Applying UniACE to 7 benchmarks across 24 domains and 15 models revealed significant score shifts, ranking reversals, and sensitivity to evidence representation, highlighting the impact of evaluation configuration on reported agent performance.
By Pengyu Zhu, Lijun Li, Yaxing Lyu, Qianxin Luo, Jingyi Yang, Yi Liu, Tingfeng Hui, Xinyu Yuan, Li Sun, Sen Su, Jing Shao
The paper discusses the problem of benchmark maintenance debt in evaluating coding agents, where benchmarks become outdated as models improve and infrastructure changes. It introduces a continuous‑validation workflow that incorporates community feedback, automated checks, and human repairs, applied to Terminal‑Bench 2.0, resulting in a revised version (Terminal‑Bench 2.1) that improves accuracy by over 10 percentage points and shifts leaderboard positions. The study also shows that resource allocation, such as time limits, can significantly alter agent performance, underscoring the need to validate both tasks and resource budgets.