{"topic_id":2708283,"version_number":1,"captured_at":"2026-08-05T10:58:14.348Z","is_encrypted":false,"main_post":{"post_number":1,"reply_to_post_number":null,"author_username":"LittleFisherman","created_at":"2026-08-05T08:23:44.183Z","content_html":"<p>glm max v2套餐<br>\n全程用5.2 max强度 自从改版后虽然说老用户不会有影响 但是体感发现消耗的速度远比以前快<br>\n5个满额五小时等于一个周额度<br>\nOpenAI 20x套餐<br>\n全部sol叠加部分terr<br>\n降价以来 用那两个模型确实量大管饱 对应的sol全部用的话可能就不是太好了 额度是肯定降了的 重置仙人没来也没有重置卡就很难受<br>\nCC 20x套餐<br>\n被封了麻了 靠 还得找渠道去用4.6<br>\nkimi 699套餐<br>\n真的只能作为临时大脑使用，相当金贵 周额度和五小时额度都在你追我赶<br>\nminimax和mimo 确实用不完 但是质量相当拉</p>\n<p>还有什么工具可以自定义子agent的模型的？除了zcode的 我想利用kimi调用glm，DeepSeek，4.6，sol,terra 当然openai做审核啦</p>","normalized_text":"glm max v2套餐 全程用5.2 max强度 自从改版后虽然说老用户不会有影响 但是体感发现消耗的速度远比以前快 5个满额五小时等于一个周额度 OpenAI 20x套餐 全部sol叠加部分terr 降价以来 用那两个模型确实量大管饱 对应的sol全部用的话可能就不是太好了 额度是肯定降了的 重置仙人没来也没有重置卡就很难受 CC 20x套餐 被封了麻了 靠 还得找渠道去用4.6 kimi 699套餐 真的只能作为临时大脑使用，相当金贵 周额度和五小时额度都在你追我赶 minimax和mimo 确实用不完 但是质量相当拉 还有什么工具可以自定义子agent的模型的？除了zcode的 我想利用kimi调用glm，DeepSeek，4.6，sol,terra 当然openai做审核啦"},"replies":[{"post_number":2,"reply_to_post_number":null,"author_username":"torvalds","created_at":"2026-08-05T08:26:27.810Z","content_html":"<p>config.toml中配置一下model_catalog_json就可以了<br>\n<div class=\"lightbox-wrapper\"><a class=\"lightbox\" href=\"https://img.linuxdo.win/posts/1110/1fa188168392811e96a244f1.png\" data-download-href=\"/uploads/short-url/5eVZUPDwrJAthvNsJ6pxOZ6l8NL.png?dl=1\" title=\"image\" rel=\"noopener nofollow ugc\"><img src=\"https://img.linuxdo.win/posts/1110/1fa188168392811e96a244f1.png\" alt=\"image\" data-base62-sha1=\"5eVZUPDwrJAthvNsJ6pxOZ6l8NL\" width=\"690\" height=\"397\" data-dominant-color=\"202122\"><div class=\"meta\"><svg class=\"fa d-icon d-icon-far-image svg-icon\" aria-hidden=\"true\"><use href=\"#far-image\"></use></svg><span class=\"filename\">image</span><span class=\"informations\">989×570 20.8 KB</span><svg class=\"fa d-icon d-icon-discourse-expand svg-icon\" aria-hidden=\"true\"><use href=\"#discourse-expand\"></use></svg></div></a></div><br>\n<div class=\"lightbox-wrapper\"><a class=\"lightbox\" href=\"https://img.linuxdo.win/posts/1110/b6c9654092d45b0f59634c2f.png\" data-download-href=\"/uploads/short-url/kWTLYUnU2KfBiNGuLAhPwG6tIhH.png?dl=1\" title=\"image\" rel=\"noopener nofollow ugc\"><img src=\"https://img.linuxdo.win/posts/1110/b6c9654092d45b0f59634c2f.png\" alt=\"image\" data-base62-sha1=\"kWTLYUnU2KfBiNGuLAhPwG6tIhH\" width=\"690\" height=\"219\" data-dominant-color=\"2B2C2D\"><div class=\"meta\"><svg class=\"fa d-icon d-icon-far-image svg-icon\" aria-hidden=\"true\"><use href=\"#far-image\"></use></svg><span class=\"filename\">image</span><span class=\"informations\">1008×320 13.2 KB</span><svg class=\"fa d-icon d-icon-discourse-expand svg-icon\" aria-hidden=\"true\"><use href=\"#discourse-expand\"></use></svg></div></a></div></p>","normalized_text":"config.toml中配置一下model_catalog_json就可以了 image 989×570 20.8 KB image 1008×320 13.2 KB"},{"post_number":3,"reply_to_post_number":null,"author_username":"hapvip808","created_at":"2026-08-05T08:27:45.965Z","content_html":"<p>对的 codex + model_xx_json  就可以 我就是用ds-v-flash 来开sub-agent的，你直接给codex说 它会自动帮你配置的</p>","normalized_text":"对的 codex + model_xx_json 就可以 我就是用ds-v-flash 来开sub-agent的，你直接给codex说 它会自动帮你配置的"},{"post_number":4,"reply_to_post_number":null,"author_username":"sineom","created_at":"2026-08-05T08:28:14.429Z","content_html":"<p>目前使用pi，配置多个provider，在skill中声明使用的provider和模型</p>","normalized_text":"目前使用pi，配置多个provider，在skill中声明使用的provider和模型"},{"post_number":5,"reply_to_post_number":2,"author_username":"hapvip808","created_at":"2026-08-05T08:28:40.675Z","content_html":"<p>你这个为什么还自带了价格了 在哪里设置吗？</p>","normalized_text":"你这个为什么还自带了价格了 在哪里设置吗？"},{"post_number":6,"reply_to_post_number":4,"author_username":"LittleFisherman","created_at":"2026-08-05T08:29:16.090Z","content_html":"<p>对应的pi agent我真玩不懂<img src=\"https://img.linuxdo.win/posts/1110/ad1dcb074a55b2b3bacd1744.png\" title=\":smiling_face_with_tear:\" class=\"emoji\" alt=\":smiling_face_with_tear:\" loading=\"lazy\" width=\"20\" height=\"20\"> 太难了 经常看到的是openai或者a​<img src=\"https://img.linuxdo.win/posts/1110/742126ea98a0de571103c5cb.png\" title=\":divide:\" class=\"emoji\" alt=\":divide:\" loading=\"lazy\" width=\"20\" height=\"20\">家的harness</p>","normalized_text":"对应的pi agent我真玩不懂 太难了 经常看到的是openai或者a​ 家的harness"},{"post_number":7,"reply_to_post_number":2,"author_username":"LittleFisherman","created_at":"2026-08-05T08:29:57.482Z","content_html":"<p>大佬 你太强了 不过这样会不会一直让sol做主导吗？ 现在sol的发散性好离谱了</p>","normalized_text":"大佬 你太强了 不过这样会不会一直让sol做主导吗？ 现在sol的发散性好离谱了"},{"post_number":8,"reply_to_post_number":6,"author_username":"xrall","created_at":"2026-08-05T08:31:41.781Z","content_html":"<p>其实自己玩不动没关系，可以让AI帮你配，就好比现在很多开源的项目也能这么操作。<br>\n当然不放心可以让他用最直白最直接巴拉巴拉的大白话告诉你</p>","normalized_text":"其实自己玩不动没关系，可以让AI帮你配，就好比现在很多开源的项目也能这么操作。 当然不放心可以让他用最直白最直接巴拉巴拉的大白话告诉你"},{"post_number":9,"reply_to_post_number":7,"author_username":"torvalds","created_at":"2026-08-05T08:32:40.737Z","content_html":"<p>肯定要选择一个强的模型做主导，你怕发散就降低推理强度，我一直sol-max用了百亿token了，感觉没什么问题</p>\n<p>别的模型作为子代理就行了，不需要/model切来切去</p>","normalized_text":"肯定要选择一个强的模型做主导，你怕发散就降低推理强度，我一直sol-max用了百亿token了，感觉没什么问题 别的模型作为子代理就行了，不需要/model切来切去"},{"post_number":10,"reply_to_post_number":9,"author_username":"LittleFisherman","created_at":"2026-08-05T08:39:02.207Z","content_html":"<p>那大佬你是怎么规划你的那些模型的？<br>\nsol作为主模型调度审核 kimi做前端 sol做后端 glm minimax和DeepSeek 打杂吗？</p>","normalized_text":"那大佬你是怎么规划你的那些模型的？ sol作为主模型调度审核 kimi做前端 sol做后端 glm minimax和DeepSeek 打杂吗？"},{"post_number":11,"reply_to_post_number":10,"author_username":"torvalds","created_at":"2026-08-05T08:44:12.885Z","content_html":"<p>差不多，不过我会随时调整子代理模型，也可以看下面那种精细化的方式</p>\n<p><div class=\"lightbox-wrapper\"><a class=\"lightbox\" href=\"https://img.linuxdo.win/posts/1110/93735bd039df5984549f48ca.png\" data-download-href=\"/uploads/short-url/5I0rIrSmmEJqIL0kUKzwAOP7nYg.png?dl=1\" title=\"image\" rel=\"noopener nofollow ugc\"><img src=\"https://img.linuxdo.win/posts/1110/afe37b88d1f9fbf4897dbd99.png\" alt=\"image\" data-base62-sha1=\"5I0rIrSmmEJqIL0kUKzwAOP7nYg\" width=\"690\" height=\"143\" srcset=\"https://img.linuxdo.win/posts/1110/afe37b88d1f9fbf4897dbd99.png, https://cdn3.ldstatic.com/optimized/4X/2/8/0/28045cfe3c588eb3820adfad141a76b31c7489ac_2_1035x214.png 1.5x, https://cdn3.ldstatic.com/optimized/4X/2/8/0/28045cfe3c588eb3820adfad141a76b31c7489ac_2_1380x286.png 2x\" data-dominant-color=\"151515\"><div class=\"meta\"><svg class=\"fa d-icon d-icon-far-image svg-icon\" aria-hidden=\"true\"><use href=\"#far-image\"></use></svg><span class=\"filename\">image</span><span class=\"informations\">1897×395 17.4 KB</span><svg class=\"fa d-icon d-icon-discourse-expand svg-icon\" aria-hidden=\"true\"><use href=\"#discourse-expand\"></use></svg></div></a></div></p>\n<p>精细化的方式，可以自定义子代理</p>\n<aside class=\"onebox allowlistedgeneric\" data-onebox-src=\"https://learn.chatgpt.com/docs/agent-configuration/subagents?surface=cli#cli-custom-agents\">\n  <header class=\"source\">\n      <img src=\"https://img.linuxdo.win/posts/1110/2b6891839e97f046af5f7e23.png\" class=\"site-icon\" alt=\"\" data-dominant-color=\"399CF9\" width=\"48\" height=\"48\">\n\n      <a href=\"https://learn.chatgpt.com/docs/agent-configuration/subagents?surface=cli#cli-custom-agents\" target=\"_blank\" rel=\"noopener nofollow ugc\">ChatGPT Learn</a>\n  </header>\n\n  <article class=\"onebox-body\">\n    <div class=\"aspect-image\" style=\"--aspect-ratio:690/361;\"><img src=\"https://img.linuxdo.win/posts/1110/2f211788e3662112015b46e6.jpg\" class=\"thumbnail\" alt=\"\" data-dominant-color=\"3B47BC\" width=\"690\" height=\"362\"></div>\n\n<h3><a href=\"https://learn.chatgpt.com/docs/agent-configuration/subagents?surface=cli#cli-custom-agents\" target=\"_blank\" rel=\"noopener nofollow ugc\">Subagents | ChatGPT Learn</a></h3>\n\n  <p>Use subagents in ChatGPT and Codex, and configure custom Codex agents</p>\n\n\n  </article>\n\n  <div class=\"onebox-metadata\">\n    \n    \n  </div>\n\n  <div style=\"clear: both\"></div>\n</aside>\n","normalized_text":"差不多，不过我会随时调整子代理模型，也可以看下面那种精细化的方式 image 1897×395 17.4 KB 精细化的方式，可以自定义子代理 ChatGPT Learn Subagents | ChatGPT Learn Use subagents in ChatGPT and Codex, and configure custom Codex agents"},{"post_number":12,"reply_to_post_number":11,"author_username":"LittleFisherman","created_at":"2026-08-05T08:46:00.350Z","content_html":"<p>好的 感激不尽<img src=\"https://img.linuxdo.win/posts/1110/4e21d32110b6e4f9b5529f0e.png\" title=\":folded_hands:\" class=\"emoji\" alt=\":folded_hands:\" loading=\"lazy\" width=\"20\" height=\"20\">  谢谢大佬啦！</p>","normalized_text":"好的 感激不尽 谢谢大佬啦！"}]}