跳到主内容
@wquguru
精选85VoxAI 产品与模型

用 OpenAI Codex 自动扫描并迁移 gpt-5.5 废弃配置

Your Codex setup may still have gpt-5.5 hardcoded in a few places. On October 14…

原文
发到 X
推荐理由

Prompt 结构非常清晰,直接给出了可复制执行的指令,详细说明了搜索范围、替换逻辑和输出格式,对开发者有直接的实操参考价值。

Your Codex setup may still have gpt-5.5 hardcoded in a few places. On October 14, 𝟭𝟭 𝗱𝗮𝘆𝘀 𝗳𝗿𝗼𝗺 𝗻𝗼𝘄, it retires from ChatGPT and Codex, and every custom agent, scheduled task and script that names it needs a new model.

你的 Codex 设置中可能仍有几处硬编码了 gpt-5.5。11月14日,即从现在起𝟭𝟭 𝗱𝗮𝘆𝘀 𝗳𝗿𝗼𝗺 𝗻𝗼𝘄(11天),它将从 ChatGPT 和 Codex 退役,所有命名它的自定义代理、定时任务和脚本都需要更换新模型。

Have Codex find them all in one pass, using OpenAI's own instructions. OpenAI also says what to switch to: GPT-6.1 Sol for complex coding and agent work, with 𝗻𝗲𝗮𝗿-𝗔𝘀𝘁𝗿𝗮 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 at a lower cost than Astra, and Luna for focused, repeatable tasks. Start with the default reasoning effort. Most tasks don't need Max or Ultra.

使用 OpenAI 自己的说明,让 Codex 一次性找出所有相关项。OpenAI 还建议切换到的目标:对于复杂的编码和代理工作,使用 GPT-6.1 Sol,其性能接近𝗻𝗲𝗮𝗿-𝗔𝘀𝘁𝗿𝗮 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲(近 Astra 水平),但成本低于 Astra;对于专注且可重复的任务,使用 Luna。从默认的推理强度开始。大多数任务不需要 Max 或 Ultra。

Send this prompt to Codex 👇

将此提示发送给 Codex 👇

"Model being retired: gpt-5.5

"正在退役的模型:gpt-5.5

Read this page: https://learn.chatgpt.com/docs/models

阅读此页面:https://learn.chatgpt.com/docs/models

Then use rg to search for that model name in this repo, in ~/.codex/ (skip chat history, logs and worktrees), and in shell config like ~/.zshrc. Only count config, code and scripts that select a model. Skip docs and notes that just mention the name. Find three kinds of problems: 1. Places where the model is hardcoded. Switch them the way OpenAI recommends: for complex coding and agent work, use gpt-6.1-sol if my account has it, otherwise gpt-6-sol. For focused, repeatable tasks, use gpt-6-luna. On Free and Go plans, use gpt-6-luna. If you're not sure which one I can use, flag it and I'll check in /model. Direct OpenAI API calls aren't affected by this retirement, so list them separately and don't count them as changes. 2. Reasoning effort that doesn't fit the work. OpenAI suggests starting with the default effort and raising it when the task needs deeper planning, and says most tasks don't need Max or Ultra. Flag anything that defaults to max or ultra. 3. Places you can't check locally: scheduled tasks in the app, workspace defaults and managed configurations. List them and tell me where to change each one by hand.

然后使用 rg 在此仓库、~/.codex/(跳过聊天记录、日志和工作树)以及 ~/.zshrc 等 shell 配置文件中搜索该模型名称。仅统计选择模型的配置文件、代码和脚本。跳过仅提及名称的文档和笔记。查找三类问题: 1. 模型被硬编码的位置。按照 OpenAI 推荐的方式进行切换:对于复杂的编码和代理工作,如果我的账户有 gpt-6.1-sol 则使用它,否则使用 gpt-6-sol。对于专注且可重复的任务,使用 gpt-6-luna。在 Free 和 Go 计划中,使用 gpt-6-luna。如果你不确定我可以使用哪一个,请标记出来,我会在 /model 中检查。直接调用 OpenAI API 不受此次退役影响,因此单独列出它们,不要将其计入更改。 2. 与工作内容不匹配的推理强度。OpenAI 建议从默认强度开始,并在任务需要更深入规划时提高强度,并指出大多数任务不需要 Max 或 Ultra。标记任何默认为 max 或 ultra 的情况。 3. 无法在本地检查的位置:应用中的定时任务、工作区默认设置和管理配置。列出它们并告诉我如何手动更改每一项。

Put everything in a table: location (file:line) | now | change to | why. Show me first, and change nothing until I confirm. When you're done, search again. Apart from the API calls, nothing should be left."

将所有内容放入一个表格中:位置(文件:行号)| 当前值 | 更改为 | 原因。先展示给我看,在我确认之前不要进行任何更改。完成后,再次进行搜索。除了 API 调用外,不应再有任何遗漏。"

更进一步:量化金融体系

看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力

进入量化体系 →

相似阅读

关联信息,但可能不是同一事件