---
id: 20261027-T0-11
title: "基准测试意外适合压缩？长任务agent的上下文压缩损耗被高估"
title_en: "Benchmarks Are Surprisingly Compaction-Friendly, Study Finds"
url: https://ai.daily.yangsir.net/daily/20261027-T0-11
issue_date: 2026-10-27
publish_date: 2026-10-26T04:00:00.000Z
category: research
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2610.31778
---

# 基准测试意外适合压缩？长任务agent的上下文压缩损耗被高估

长周期agent任务正把LLM agent推过上下文窗口上限，agent框架必须腾出上下文空间，这一过程叫上下文压缩（context compaction）。压缩按设计通常是有损的，agent可能悄悄丢掉关键信息。这篇论文考察了基准测试在压缩下的表现，发现基准测试对压缩的容忍度出乎意料地高。这提示现有评测可能低估了压缩的实际风险，也说明用基准分数判断长任务agent的可靠性存在盲区。

## English Version

**Benchmarks Are Surprisingly Compaction-Friendly, Study Finds**

Long-horizon agentic tasks are pushing LLM agents past their context windows, forcing agent scaffolds to free up space through context compaction. Compaction is generally lossy by design, and agents can silently drop critical information. This paper examines how benchmarks behave under compaction and finds they are surprisingly tolerant of it. The result suggests existing evaluations may understate compaction's real risks, and that benchmark scores leave a blind spot when judging long-task agent reliability.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2610.31778)

**详情页**：https://ai.daily.yangsir.net/daily/20261027-T0-11

---

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