Saved in:
Bibliographic Details
Main Authors: Yang, Qianyu, Liu, Yang, Li, Jiaqi, Bai, Jun, Chen, Hao, Chen, Kaiyuan, Duan, Tiliang, Dong, Jiayun, Hu, Xiaobo, Jia, Zixia, Peng, Tao, Ren, Yixin, Tian, Ran, Wang, Zaiyuan, Xiao, Yanglihong, Yao, Gang, Yin, Lingyue, Zhang, Ge, Zhang, Chun, Jiao, Jianpeng, Zheng, Zilong, Gong, Yuan
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.07980
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908873128935424
author Yang, Qianyu
Liu, Yang
Li, Jiaqi
Bai, Jun
Chen, Hao
Chen, Kaiyuan
Duan, Tiliang
Dong, Jiayun
Hu, Xiaobo
Jia, Zixia
Liu, Yang
Peng, Tao
Ren, Yixin
Tian, Ran
Wang, Zaiyuan
Xiao, Yanglihong
Yao, Gang
Yin, Lingyue
Zhang, Ge
Zhang, Chun
Jiao, Jianpeng
Zheng, Zilong
Gong, Yuan
author_facet Yang, Qianyu
Liu, Yang
Li, Jiaqi
Bai, Jun
Chen, Hao
Chen, Kaiyuan
Duan, Tiliang
Dong, Jiayun
Hu, Xiaobo
Jia, Zixia
Liu, Yang
Peng, Tao
Ren, Yixin
Tian, Ran
Wang, Zaiyuan
Xiao, Yanglihong
Yao, Gang
Yin, Lingyue
Zhang, Ge
Zhang, Chun
Jiao, Jianpeng
Zheng, Zilong
Gong, Yuan
contents As language models (LMs) evolve from chat assistants to long-horizon agents capable of multi-step reasoning and tool use, existing benchmarks remain largely confined to structured or exam-style tasks that fall short of real-world professional demands. To this end, we introduce \$OneMillion-Bench \$OneMillion-Bench, a benchmark of 400 expert-curated tasks spanning Law, Finance, Industry, Healthcare, and Natural Science, built to evaluate agents across economically consequential scenarios. Unlike prior work, the benchmark requires retrieving authoritative sources, resolving conflicting evidence, applying domain-specific rules, and making constraint decisions, where correctness depends as much on the reasoning process as the final answer. We adopt a rubric-based evaluation protocol scoring factual accuracy, logical coherence, practical feasibility, and professional compliance, focused on expert-level problems to ensure meaningful differentiation across agents. Together, \$OneMillion-Bench provides a unified testbed for assessing agentic reliability, professional depth, and practical readiness in domain-intensive scenarios.
format Preprint
id arxiv_https___arxiv_org_abs_2603_07980
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle \$OneMillion-Bench: How Far are Language Agents from Human Experts?
Yang, Qianyu
Liu, Yang
Li, Jiaqi
Bai, Jun
Chen, Hao
Chen, Kaiyuan
Duan, Tiliang
Dong, Jiayun
Hu, Xiaobo
Jia, Zixia
Liu, Yang
Peng, Tao
Ren, Yixin
Tian, Ran
Wang, Zaiyuan
Xiao, Yanglihong
Yao, Gang
Yin, Lingyue
Zhang, Ge
Zhang, Chun
Jiao, Jianpeng
Zheng, Zilong
Gong, Yuan
Machine Learning
Artificial Intelligence
Computation and Language
As language models (LMs) evolve from chat assistants to long-horizon agents capable of multi-step reasoning and tool use, existing benchmarks remain largely confined to structured or exam-style tasks that fall short of real-world professional demands. To this end, we introduce \$OneMillion-Bench \$OneMillion-Bench, a benchmark of 400 expert-curated tasks spanning Law, Finance, Industry, Healthcare, and Natural Science, built to evaluate agents across economically consequential scenarios. Unlike prior work, the benchmark requires retrieving authoritative sources, resolving conflicting evidence, applying domain-specific rules, and making constraint decisions, where correctness depends as much on the reasoning process as the final answer. We adopt a rubric-based evaluation protocol scoring factual accuracy, logical coherence, practical feasibility, and professional compliance, focused on expert-level problems to ensure meaningful differentiation across agents. Together, \$OneMillion-Bench provides a unified testbed for assessing agentic reliability, professional depth, and practical readiness in domain-intensive scenarios.
title \$OneMillion-Bench: How Far are Language Agents from Human Experts?
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2603.07980