4-5 days replacing Claude w Qwen 3.8 Next

Hi, human here with rambling thoughts to share. Feel free to skip Overall, I don’t feel like I’m missing much; if anything. On my hardware(M1 ultra w 128Gb) it’s probably not as fast as Claude but I’ve been using opencode for research and other business related tasks and it’s been getting the job done and learning. I might dive back in for the multi agent workflows that speed things up with a cloud provider but I’m working on setting up different slots as I refine my custom harness which works well for chat but not all the actual fun and useful stuff. Claude “knew” me better but that’s to be expected after months of back and forth with it and I’m honestly not sure I want them to know me this well. Not expecting a lot of responses but it’s pretty cool I’m able to replace the service a billion dollar company provides with a Mac Studio and free software. Any advice on better optimizing my system to improve speed without sacrificing accuracy? Planning to work on optimizing DeepSeek V4 0731 and GLM Flash but they don’t seem to be “better” than Qwen 3.8 next so i decided to start spending more time using instead of optimizing for prefill and tokens per second.