<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>评测基准 on LoopRouter Blog</title><link>https://blog.looprouter.com/tags/%E8%AF%84%E6%B5%8B%E5%9F%BA%E5%87%86/</link><description>Recent content in 评测基准 on LoopRouter Blog</description><generator>Hugo</generator><language>zh-cn</language><lastBuildDate>Thu, 17 Sep 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://blog.looprouter.com/tags/%E8%AF%84%E6%B5%8B%E5%9F%BA%E5%87%86/index.xml" rel="self" type="application/rss+xml"/><item><title>评估一个 Agent 模型要烧 1 万美元、跑 2.7 天:20 个任务的迷你集把选型成本压到 1/40</title><link>https://blog.looprouter.com/docs/agent-eval-cost-miniset-model-selection/</link><pubDate>Thu, 17 Sep 2026 00:00:00 +0000</pubDate><guid>https://blog.looprouter.com/docs/agent-eval-cost-miniset-model-selection/</guid><description>&lt;blockquote&gt;&#10;&lt;p&gt;本文基于 arXiv 论文《Beyond Outcomes: Dual-View Relational Learning for Efficient Agent Benchmarking》(DualViewEval)的公开内容整理与二次分析,加入 LoopRouter 的选型/路由视角;原文见文末链接。&lt;/p&gt;</description></item><item><title>选模型别只看跑分:证据一「埋深」,每答对一题的成本翻了 4.7–7 倍</title><link>https://blog.looprouter.com/docs/agentic-qa-cost-per-correct-answer/</link><pubDate>Tue, 15 Sep 2026 00:00:00 +0000</pubDate><guid>https://blog.looprouter.com/docs/agentic-qa-cost-per-correct-answer/</guid><description>&lt;blockquote&gt;&#10;&lt;p&gt;本文基于 arXiv 预印本《Clean Scores, Buried Evidence, and Confident Wrong: A Receipt-Based Audit of Frontier Agentic QA》(arXiv:2609.15319v1)的公开内容整理与二次分析,加入 LoopRouter 在模型选型、成本治理与路由/故障切换上的视角;原文见文末链接。该研究为合成「数据室」上的预印本、且明确声明不做模型排名,文中数字请读作「方法示例」而非厂商座次,建议以你自己的数据复测。&lt;/p&gt;</description></item></channel></rss>