@YukitokoG Я тоже попал в бан клода и пришлось переходить на кодекс.
Я вывел теорию:
1. Кодекс юзают вайбкодеры просто потому что он не отказывается и не задумывается, поэтому тебе не приходится думать
2. Дипсики и прочее юзают реальные инженеры - пишут спеки, а дипсик чисто исполняет
@sudoasdk@facedevstuff Looking at qwen 27b, haiku seems reasonable But Opus 700b nah - more like Sonnet 700b (as DS-4.1flash is 700b? or 500b, cant remember for sure), Opus 1-2T and Fable 1-xT
multi 1.18.0 is here! multi is ai plugin that let's you run multi-ple models for different tasks - mainly for code-review, which was proven to find 2x more bugs
- kilo code as backend for more free models
- disable models that trains on u with --no-train
https://t.co/C0PVxDnIUN
multi 1.17.0 is here!
multi is ai plugin that let's you run multi-ple models for different tasks - mainly for code-review, which was proven to find 2x more bugs
new:
/multi:skill <skill-name> - run same skill with multi-ple models
github copilot support
https://t.co/C0PVxDnb5f
> new model X
> "omg its beats every other model"
> "i built <game name> in one prompt! AGI is here"
> 2 weeks passed
> "omg i hate X wtf is wrong with it. i cancel my plan"
> new model Y
> "omg it beats every other model, i built <game name> in one prompt, AGI is here>
> repeat
i can't remember what broke our database two months ago. my agent definitely can't — it forgets on every compact.
so: don't forget. a folder of .md notes your agent writes and searches. local, plaintext, stdlib only.
https://t.co/uyOnd0PQcS
#deltarune lmao
the model that wrote the code is the worst judge of whether it works.
so multi asks the ones that didn't. codex, opencode, claude subagents — same diff, three opinions, no merging.
claude plugin marketplace add szarkans/multi
https://t.co/C0PVxDnb5f