<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[Club-3090 俱樂部 頂尖對決 GLM-5.3-Flash vs Qwen3.8-Flash-Next  vs. Qwen3.8-27B]]></title><description><![CDATA[<p dir="auto"><strong>資料很多, 有重點沒提到的歡迎補上喔 <img src="https://lcz.me/assets/plugins/nodebb-plugin-emoji/emoji/android/1f60a.png?v=ec85ec55bc1" class="not-responsive emoji emoji-android emoji--blush" style="height:1.4em;width:auto;vertical-align:middle" title=":blush:" alt="😊" /></strong></p>
<h2>GPT Image 圖像快速解說版</h2>
<p dir="auto"><img src="https://upload.lcz.me/uploads/6ac93de2-596a-46d7-9ebc-5437d4534d0f.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
<blockquote></blockquote>
<h2><strong>原文作者 : paulp83</strong></h2>
<blockquote>
<p dir="auto"><strong>配置：2塊RTX 5090顯示卡，TR PRO 3975WX散熱器，256GB記憶體。大約550個代理程式運行了三天半</strong></p>
</blockquote>
<p dir="auto">3 位local LLMs 作為代理完成了以下工作：20 個 club-3090 修復、38 個桌面任務、2 個安全帶、2 個工作量級別 #1690<br />
<a href="https://github.com/noonghunna/club-3090/discussions/1690" rel="nofollow ugc">https://github.com/noonghunna/club-3090/discussions/1690</a></p>
<blockquote>
<p dir="auto">我詢問俱樂部的測試結果是否與我使用的模型和框架有關，維護者要求我做一個 2×2 的對比測試：低強度模型與高強度模型，裸代理與 OMP。我在這個倉庫中提供了三個模型，並運行了上述 2×2 測試，然後給其中兩個模型佈置了一組非編碼的桌面任務。以下是測試結果，包括每個模型的優勢和劣勢，以及我現在使用的設定。</p>
</blockquote>
<h2>受測目標模型</h2>
<p dir="auto"><img src="https://upload.lcz.me/uploads/fe0bf7cd-86d8-450e-b56b-8d32187e375a.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
<h2>受測結果 成功解題比率</h2>
<p dir="auto"><img src="https://upload.lcz.me/uploads/a6aae684-8650-43c1-9c6f-ee7548d1dfe6.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
<h2>每小時的解題效率</h2>
<p dir="auto"><img src="https://upload.lcz.me/uploads/b83937a7-bb8c-4cc4-b2a8-4c1303041f8b.jpeg" alt="image.jpeg" class="img-markdown  img-fluid" /></p>
<table class="table table-bordered table-striped">
<thead>
<tr>
<th>Config</th>
<th>Solved</th>
<th>95% CI</th>
<th>Partial credit</th>
<th>Right files</th>
<th>Median min/task</th>
<th>Solved per agent-hour</th>
</tr>
</thead>
<tbody>
<tr>
<td>Flash Next IQ4_XS</td>
<td>31/40</td>
<td>62–88%</td>
<td>0.85</td>
<td>0.93</td>
<td>9.7</td>
<td>4.3</td>
</tr>
<tr>
<td>GLM avar6</td>
<td>13/20</td>
<td>43–82%</td>
<td>0.72</td>
<td><strong>0.96</strong></td>
<td>12.6</td>
<td>2.5</td>
</tr>
<tr>
<td>Flash Next Q4_K_XL</td>
<td>25/40</td>
<td>47–76%</td>
<td>0.71</td>
<td>0.88</td>
<td>7.6</td>
<td>3.5</td>
</tr>
<tr>
<td>GLM IQ4_XS</td>
<td>12/20</td>
<td>39–78%</td>
<td>0.70</td>
<td>0.93</td>
<td>12.0</td>
<td>2.3</td>
</tr>
<tr>
<td>ThinkingCap-27B</td>
<td>23/40</td>
<td>42–71%</td>
<td>0.68</td>
<td>0.87</td>
<td><strong>4.4</strong></td>
<td><strong>7.0</strong></td>
</tr>
</tbody>
</table>
]]></description><link>https://lcz.me/topic/2338</link><generator>RSS for Node</generator><lastBuildDate>Mon, 26 Oct 2026 05:34:22 GMT</lastBuildDate><atom:link href="https://lcz.me/topic/2338.rss" rel="self" type="application/rss+xml"/><pubDate>Thu, 15 Oct 2026 10:20:46 GMT</pubDate><ttl>60</ttl><item><title><![CDATA[Reply to Club-3090 俱樂部 頂尖對決 GLM-5.3-Flash vs Qwen3.8-Flash-Next  vs. Qwen3.8-27B on Thu, 15 Oct 2026 12:51:34 GMT]]></title><description><![CDATA[<p dir="auto">可以，以后完全不用自己写稿子，打开电脑，照着念一个帖子。</p>
]]></description><link>https://lcz.me/post/25079</link><guid isPermaLink="true">https://lcz.me/post/25079</guid><dc:creator><![CDATA[terry]]></dc:creator><pubDate>Thu, 15 Oct 2026 12:51:34 GMT</pubDate></item><item><title><![CDATA[Reply to Club-3090 俱樂部 頂尖對決 GLM-5.3-Flash vs Qwen3.8-Flash-Next  vs. Qwen3.8-27B on Thu, 15 Oct 2026 12:50:39 GMT]]></title><description><![CDATA[<p dir="auto">物理定律在那里锁死的，只能等STRATA作者深入发掘一下大内存和大带宽下面的潜力了。</p>
]]></description><link>https://lcz.me/post/25078</link><guid isPermaLink="true">https://lcz.me/post/25078</guid><dc:creator><![CDATA[stxpnet]]></dc:creator><pubDate>Thu, 15 Oct 2026 12:50:39 GMT</pubDate></item><item><title><![CDATA[Reply to Club-3090 俱樂部 頂尖對決 GLM-5.3-Flash vs Qwen3.8-Flash-Next  vs. Qwen3.8-27B on Thu, 15 Oct 2026 11:28:31 GMT]]></title><description><![CDATA[<p dir="auto"><a class="plugin-mentions-user plugin-mentions-a" href="/user/johnnybegood" aria-label="Profile: johnnybegood">@<bdi>johnnybegood</bdi></a></p>
<p dir="auto">看來還是QFN 比較香甜</p>
<p dir="auto">不過有四台DGX Spark跑高精度的GLM 說不定效果不同</p>
]]></description><link>https://lcz.me/post/25071</link><guid isPermaLink="true">https://lcz.me/post/25071</guid><dc:creator><![CDATA[kos or]]></dc:creator><pubDate>Thu, 15 Oct 2026 11:28:31 GMT</pubDate></item><item><title><![CDATA[Reply to Club-3090 俱樂部 頂尖對決 GLM-5.3-Flash vs Qwen3.8-Flash-Next  vs. Qwen3.8-27B on Thu, 15 Oct 2026 10:47:54 GMT]]></title><description><![CDATA[<p dir="auto">这些模型我也都跑过哎， 虽然没有测这么全， 但是这个结果跟我的感觉是一样的。最扯的是 GLM5.3 Flash IQ2 还会在数学题上算错，而GLM IQ4我跑起来只有 15tok/s</p>
]]></description><link>https://lcz.me/post/25063</link><guid isPermaLink="true">https://lcz.me/post/25063</guid><dc:creator><![CDATA[johnnybegood]]></dc:creator><pubDate>Thu, 15 Oct 2026 10:47:54 GMT</pubDate></item></channel></rss>