(I'm on Windows system running these models locally)
I've used both Codex and OpenCode with Qwen 3.6 27b and 35b running locally.
I'm having a bitch of a time getting them to correctly create files. Sometimes even just editing contents of a file they fall on their face.
OpenCode with Qwen 3.6 27b running with vLMM
Codex with Ollama and Qwen 3.6 35b
They continue to constantly barf whenever they need to create files.
For a simple prompt like "Create an HTML/CSS webpage for a salon , showing services, a map, and a contact page"
In Open Code it kept failing JSON formatting just to write the HTML file content
In Codex it constantly dies with complications writing correctly with powershell.
Heck even just asking it to change an address in the HTML would be faster for me to do it myself, the tool calling just loops for 1-2 minutes making JSON mistakes /etc .
In OpenCode the AI finally resorted to writing a python script to *create my HTML file*. Like wtf dude just write the text out to a file! You are literally a text machine.
At one point with Codex and Qwen 3.6 35b it kept telling me it wrote some files out / created them but they never got actually created.
Why does this AI model just have nothing but problems simply *creating a damn file*
It really seems like it has the most trouble with web files like HTML / CSS. I was able to edit a C++ file earlier today without a lot of fuss (surprisingly)
It constantly barfs at editing/creating files.
[link] [comments]