The paper introduces a disclosure‑gated user simulator for companion‑agent evaluation, where information release is conditioned on the agent’s behavior through a five‑level gate system. The authors train the simulator on synthetic and real data, audit its performance, and demonstrate that it preserves ranking order and score stability across 12 tested systems. Their released simulator achieves a 0.993 correlation with the original benchmark’s simulator and outperforms prompting a frontier model, which only shifts scores upward without affecting rankings.
By Yao Liu, Yu He
arXiv:2609.38043v1 Announce Type: new
Abstract: Interactive agent benchmarks and multi-turn reinforcement learning increasingly place a second language model in the role of the user. This simulated u...
By Ashish Jain, Armaan Sandhu
GAUGE is a new offline protocol that evaluates whether the common practice of using an LLM-as-a-judge to rank task‑oriented agents actually aligns with a verifiable reward. Across 25 agents from six providers on two benchmarks, GAUGE finds that user satisfaction scores are largely uncorrelated with task success, and that the judge’s ranking loses precision when agents are closely matched in performance. The study highlights a gap between ranking validity and construct validity in current evaluation practices.
By Umesh Bodhwani, Thanh Tran, Kai Wei
arXiv:2606. 25760v1 Announce Type: new Abstract: Computer-use agents turn vision-language model (VLM) predictions into executable GUI clicks, so reliable uncertainty estimates are essential for rejection, calibration, miss-severity ranking, and spatial safety regions.
By Divake Kumar, Sina Tayebati, Devashri Naik, Amanda Sofie Rios, Nilesh Ahuja, Omesh Tickoo, Ranganath Krishnan, Amit Ranjan Trivedi
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
SWE‑Duel proposes a self‑scaling benchmark for coding agents, where tasks and oracles are generated at evaluation time to prevent data leakage and adapt difficulty to participant performance. In the arena, two agents take turns as Red and Blue: Red proposes a feature, injects a subtle bug, and must repair it blind, while Blue reviews the pull request without knowing the bug and must pass all tests. A tournament of 11 model‑harness participants on 15 repositories across five languages showed GPT‑5.5/Codex CLI as the top solver and Claude Opus 4.8/mini‑swe‑agent as the top task author, with newer model versions producing harder tasks for older ones.
By Wei Minn, Zhensu Sun, Lwin Khin Shar, David Lo
arXiv:2610.03448v1 Announce Type: cross
Abstract: LLM agents increasingly screen tool outputs with small prompt-injection detectors, and teams choose among detectors by their scores on public benchma...
By Zhuowen Liu
arXiv:2604. 16706v2 Announce Type: replace Abstract: Automated evaluation of tool-using large language model (LLM) agents is widely assumed to be reliable, yet this assumption is rarely validated against human annotation.
By Bhaskar Gurram
The paper introduces Conservative Meta-Agent Search (CMAS), a method for adapting agent architectures in non‑stationary environments. CMAS retrieves archived designs, proposes typed graph edits, and evaluates them through paired shadow replay with variance‑penalized scoring and a simulated canary stage before promotion. Using the MetaInvest‑Bench simulator, CMAS reduces surrogate dynamic‑oracle regret and improves held‑out utility across 96 financial streams, while maintaining low false‑promotion rates and modest inference cost.
By Ali Atiah Alzahrani
SWE‑Duel proposes a self‑scaling benchmark for coding agents, where tasks and oracles are generated at evaluation time to prevent data leakage and adapt difficulty to the agents’ capabilities. In the arena, two agents take turns modifying a real repository: the Red agent adds a feature, introduces a subtle bug, and the Blue agent must fix it without knowing the bug exists. A tournament of 11 model‑harness participants on 15 repositories in five languages showed GPT‑5.5 as the best solver and Claude Opus 4.8 as the best task author, with newer model versions producing harder tasks for older ones.
arXiv:2606. 03650v1 Announce Type: cross Abstract: Choosing or ranking language models for a specific application is hardest when no task-specific labeled data exists, and standard public benchmarks cannot be trusted, their items having likely leaked into pretraining, so scores reflect memorization rather than fitness.
By Alexander Apartsin, Yehudit Aperstein
arXiv:2606. 14199v1 Announce Type: cross Abstract: Large language models are increasingly deployed as human simulators for interactive evaluation and social simulation.
By Xuhui Zhou, Weiwei Sun, Weihua Du, Jiarui Liu, Haojia Sun, Qianou Ma, Tongshuang Wu, Yiming Yang, Maarten Sap