What do you guys use local models for?
这条记录涉及编程工具或代码能力更新，适合开发者评估工作流变化和可复用价值。
用户询问大家如何使用本地模型。他曾开发一个Telegram聊天机器人，但未投入生产。他发现即使是ChatGPT 4.1mini等廉价高级模型也“太笨”，无法很好地理解任务，需要多次调用才能完成，但效果不佳。因此，他好奇7B或35B模型能用于哪些工作流程，例如是否有人用它们编写代码。…
I previously built a telegram bot that chatted with people. But it was never put in production.
Besides that I often find that even the cheap tier premium models like ChatGPT 4.1mini are a little bit too stupid to do things. They just don't seem to grasp the job as well as the bigger models. You have to compartmentalize everything with separate calls but even that doesn't always work.
So I'm just curious what kind of work flows can be done with these 7B or 35B models. Do you guys write code? I've been using mostly Claude Code with Sonnet 5 (and Opus 4.6 prior to that) and Codex with 5.5. Again since those models don't always perform I can't imagine what a 7B model would do.