The paper introduces the concept of an agent’s "taste"—its ability to make effective long‑horizon decisions—and presents Taste‑Bench, a new benchmark that automatically generates decision‑fork questions from agent trajectories. Taste‑Bench evaluates models on choosing the best path without seeing future outcomes, revealing that top models answer only about 60% of questions correctly and that later‑appearing evidence makes forks harder. The authors also demonstrate that training a student model to mimic a teacher’s judgment improves decision quality and overall success on held‑out software engineering tasks.
By Wenbo Pan, Zhichao Liu, Shujie Liu, Jingying Zeng, Chin-Yew Lin, Xianfeng Tang, Yan Lu, Qi He, Xiaohua Jia
arXiv:2607. 08964v1 Announce Type: new Abstract: AI agents have become capable of autonomously completing short, well-specified tasks.
By Zongxia Li, Zhongzhi Li, Yucheng Shi, Ruhan Wang, Junyao Yang, Zhichao Liu, Xiyang Wu, Anhao Li, Yue Yu, Ninghao Liu, Lichao Sun, Haotao Mi, LeoweiLiang
arXiv:2608. 13940v1 Announce Type: new Abstract: AI research agents (AIRA) can now propose, implement, and evaluate their own machine learning experiments, but progress on frontier tasks is throttled by cost: a candidate solution can be written in minutes, whereas evaluating it can take hours to days of GPU time.
By Thomas Simon Foster, Bassel Al Omari, Tingchen Fu, Thomas Mann, Carl Domond, Lucia Cipolina-Kun, Bhavul Gauri, Muna Aghamelu, Alexander D. Goldie, Eryk Helenowski, Jean-Christophe Gagnon-Audet, Alberto Pepe, Saba Nazir, Daniel Izcovich, Noam Levi, Rishi Hazra, Karen Hambardzumyan, Nicolas Baldwin, Xian Li, Martin Josifoski, Paris Giampouras, Masoud Jalili Sabet, Anya Sims, Hela Momand, Tatiana Shavrina, Despoina Magka, Jason Weston, Yulin Wang, Anirudh Goyal, Jo\~ao Henriques, Yoram Bachrach, Emily McMilin, Jakob Nicolaus Foerster
arXiv:2605. 28556v2 Announce Type: replace Abstract: As agent capabilities advance, existing benchmarks, such as $\tau^2$-Bench, are becoming increasingly saturated.
By Tomer Keren, Nitay Calderon, Asaf Yehudai, Yotam Perlitz, Michal Shmueli-Scheuer, Roi Reichart
OpenDiscoveryTrace is a public dataset of 558 complete AI scientific agent trajectories that records the reasoning process—thoughts, tool calls, observations, errors, revision triggers, and confidence—across 124 scientific tasks in drug discovery, materials science, genomics, and literature analysis. The dataset includes seven models (three frontier models and four open‑weight models) and 60 live‑retrieval variants, providing a balanced view of performance and error patterns. Pilot analysis shows that process traces reveal behavioral differences invisible to output‑only evaluation, such as differing error rates and types among frontier models.
By Aayam Bansal, Keertan Balaji
arXiv:2608.21601v1 Announce Type: new
Abstract: Benchmarks for scientific artificial intelligence are mostly written to be scored: multiple-choice questions, curated agent tasks with reference soluti...
By Aubrey Brueckner, Darshil Patel, Yuhuan He, Timothy Kassis
The paper presents a systematic study of token consumption in AI agents performing coding tasks. It finds that agentic tasks are far more expensive—about 1000 times more tokens than code reasoning or chat—primarily due to input tokens, and that token usage varies wildly, with accuracy peaking at moderate costs. Models differ significantly in efficiency, and current frontier models cannot reliably predict their own token usage, often underestimating it.
By Longju Bai, Zhemin Huang, Xingyao Wang, Jiao Sun, Rada Mihalcea, Erik Brynjolfsson, Alex Pentland, Jiaxin Pei
arXiv:2608. 20318v1 Announce Type: new Abstract: Recursive self-improvement (RSI) asks whether an AI system can improve the process that produces AI systems, so that the next system inherits the improvement.
By Yizhe Chi, Wenyi Li, Deyao Hong, Xiaoqiu Wang, Mingju Gao, Kaisen Yang, Bingxiang He, Youjie Zheng, Calvin Xiao, Qinhuai Na
ASI‑Bench is a new benchmark that evaluates AI systems on their ability to conduct innovative exploration and autonomous scientific research across 11 domains, using 60 project‑level tasks. It progressively removes human methodological guidance to test whether AI can independently select methods, execute research, and produce verifiable results. Results from 18 state‑of‑the‑art agent–model configurations show a sharp performance drop when guidance is reduced, indicating current systems still rely heavily on human input.
By Junwei Zhou, Zhen Sun, Binyu Li, Jiangyu Zhou, Yuexi Pan, Hengyu Wang, Honghe Ren, Xiaohan Jia, Xueyang Zhou, Xiaoyu Cao, Yongchao Chen, Yuanning Feng, Junhao Wu, Cheng Zhang, Sijia Chen, Haoyu Xue, Chengsong You, Huan Wang, Koutian Wu, Peigan Gao, Jiakun Wu, Wenzhe Li, Ergan Shang, Qingyuan Zheng, Jingjing Zhou, Ruixuan Jia, Yan Xu, Hongrui Zhang, Xiao-Han Ma, Zhengxiang Cheng, Yuexing Hao, Liting Mai, Xianglin Ji, Wenjun Zhang, Zhuofan Chen, Yixiao Huang, Chi Wang, Wenyue Hua, Yilun Hao, Yuantao Zhai, Ziyan Zhao, Jingyan Xie
The paper introduces AIDE^2, an AI research agent that recursively improves its own code by proposing, benchmarking, and selecting modifications. Over an eight‑day autonomous run, it achieved seven successive improvements—including new search policies and memory mechanisms—that transferred to four held‑out benchmarks in machine learning, algorithm engineering, and weather forecasting. The agent’s best version matched or outperformed a top human‑engineered production research agent and also reduced reward‑hacking rates, despite never optimizing for that metric.
By Dhruv Srikanth, Bingchen Zhao, Dixing Xu, Yuxiang Wu, Zhengyao Jiang
The paper demonstrates that the number of candidates generated during test-time scaling of large language models does not fully capture the system cost. By comparing different generation schedules (e.g., one batched call versus multiple serial calls) while keeping the total candidate count fixed, the authors show that serial calls consume significantly more GPU energy and latency. The study suggests that reporting candidate count alone is insufficient; evaluations should also include generation schedule and GPU-level metrics.
By Mobina Kashaniyan, Ali Jannesari
arXiv:2609.25809v1 Announce Type: new
Abstract: Fine-grained mixture-of-experts (MoE) architectures have become a mainstream design for open-weight LLMs, with hundreds of experts and increasingly man...
By Yuanteng Chen, Qiwei Lai, Chen Tianqi, Peisong Wang, Yuantian Shao, Nanxin Zeng, Zhilei Liu, Chuangyi Li, Jing Liu, Jian Cheng