<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Π0.5 on ElephantFlow's Blog</title><link>https://elephantflow.github.io/tags/%CF%800.5/</link><description>Recent content in Π0.5 on ElephantFlow's Blog</description><generator>Hugo -- 0.146.0</generator><language>zh-CN</language><lastBuildDate>Wed, 16 Sep 2026 09:00:00 +0800</lastBuildDate><atom:link href="https://elephantflow.github.io/tags/%CF%800.5/index.xml" rel="self" type="application/rss+xml"/><item><title>GPT-6 屠榜？具身智能还能做什么</title><link>https://elephantflow.github.io/clips/gpt-6-astra-benchmark-what-remains/</link><pubDate>Wed, 16 Sep 2026 09:00:00 +0800</pubDate><guid>https://elephantflow.github.io/clips/gpt-6-astra-benchmark-what-remains/</guid><description>一篇匿名评测显示 GPT-6 Astra 不微调就能直接控机械臂，在 RoboLab 上 49/50（98%）压过 π0.5 与混合策略。但把任务换成柔性物体就反转了——叠衣服 Direct 40%、混合 100%，且 Direct 按成功任务分摊的 Token 是混合的 3.35 倍。作者的…</description></item></channel></rss>