WHAT NEW IN X10.28 WED

AI collections from X · 22 collections · 58,925 posts

New todaydeepseek-v4.1POST

GPT-6.1 Sol plans, DeepSeek 1017 executes: a working split

A community workflow uses GPT-6.1 Sol to write specs and accept results while DeepSeek 1017 does the work, skipping second reviews to avoid rework and wasted tokens.

明子206
@mingzi206Archived 9 hours agoOriginally posted 2026.10.27
明子206 @mingzi206Originally posted 2026.10.27
Open original on X

拿 GPT-6.1 Sol 当主模型规划和审查、DeepSeek 1017 当子模型干活的搭配,社区已经踩出一套能用的流程了 明子把回复里的坑捋了一遍,审查千万别让模型审自己,DeepSeek 审 DeepSeek 等于没审,Sol 审的时候范围也要收住,不然它会跑偏去做计划外的测试 · 主模型定好 spec 再派活,子模型做完由主模型收尾验收,别再二次 review,不然会没完没了打回重做 · Sol 进 goal 模式等子代理时可能空跑烧 Token,实测有人加个注释开了子代理干了 40 分钟 分工本身没问题,慢和返工才是成本,流程定死比换模型管用 你现在的主模型和子模型是怎么搭的?

x.com/mingzi206/status/2115057423312621907Archived 2026.10.28 12:05