有人在32GB Mac上使用Qwen3.6-35B-A3B-UD-Q4_K_M透過opencode、claude code或類似工具完成實際編碼工作嗎？

I'm running Qwen3.6-35B-A3B-UD-Q4_K_M on an M2 Macbook Pro with 32GB of RAM. I'm using quite recent builds of llama.cpp and opencode. To avoid llama-server crashing outright due to memory exhaustion, I have to set the context window to 32768 tokens. This turns out to be important. As a hopefully reasonable test, I gave opencode a task that Claude Code was previously able to complete with Opus 4.7. The project isn't huge, but the task involves rooting around the front and back end of an applica