Am I Expecting Too Much?
I’ve been trying to get started with local coding help after finding Claude Code really useful but I’m just running into a ton of problems.
On my RTX 5090, I’ve been running Qwen 3.6 27B UD_4 at 131K context, no KV quants, quant’d by Unsloth. I’ve been using Cline in VS Code, as it seemed the closest parallel to Claude Code.
Recently, I wanted to try implementing some plans I had made with Fable. I had Fable prepare detailed, junior-engineer-friendly implementation plans for a pretty basic Python app, which I threw into the repo.
But I just can’t get anywhere with Qwen. It writes code that contains a ton of mistakes, tries to write terminal commands that are just broken with basic syntax errors, etc. It can’t follow even these very detailed plans going basically step by step.
To be clear, I’m not expecting Opus level performance or even Sonnet level work, but it just doesn’t work at all, even within the harness and via basic terminal calls.
Am I expecting too much?
Is something in my approach or setup wrong?
Is there a better harness or model I should swap to?