---
id: 20261018-T0-03
title: "实时环境下Agent动作会“迟到”：RealtimeBench用180个任务测LLM代理的时间敏感性"
title_en: "RealtimeBench Tests LLM Agents on a Live Clock With 180 Time-Sensitive Tasks"
url: https://ai.daily.yangsir.net/daily/20261018-T0-03
issue_date: 2026-10-18
publish_date: 2026-10-17T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2610.20665
---

# 实时环境下Agent动作会“迟到”：RealtimeBench用180个任务测LLM代理的时间敏感性

研究指出，LLM代理能完成复杂交互任务，但推理和工具调用都要花时间，在持续变化的环境中，本来正确的动作可能因为到达太晚而失效。为此作者提出RealtimeBench，包含六个类别共180个任务，在实时时钟下评估代理。这补上了现有基准普遍忽略的一环：动作的正确性不只取决于内容，还取决于时机。对做代理部署的团队，这意味着评估体系需要把延迟和时效纳入指标，否则离线测试的高分未必能转化为线上可用性。

## English Version

**RealtimeBench Tests LLM Agents on a Live Clock With 180 Time-Sensitive Tasks**

LLM agents can solve complex interactive tasks, but reasoning and tool calls take time, so in continuously changing environments an otherwise correct action may arrive too late. The authors introduce RealtimeBench, with 180 tasks across six categories, to evaluate agents under a live clock. It fills a gap most existing benchmarks ignore: correctness depends not just on what an agent does but on when it does it. For teams deploying agents, this means evaluation needs to treat latency and timeliness as first-class metrics, since strong offline scores may not translate into usable online behavior.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2610.20665)

**详情页**：https://ai.daily.yangsir.net/daily/20261018-T0-03

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*