I have used Sol Ultra + Fast mode 2-3 times in the past two weeks, and my 24GB M4 Pro ran out of memory every single time without exception.
It is very obvious that it does not know when to stop spawning new workers, which causes my Mac to become completely frozen. The only way to recover is to restart the machine.
Also, the quota consumption is extremely fast, but it still cannot complete tasks well. I found that extra high thinking actually performs much better than Ultra.
Regarding what I mentioned two weeks ago about frontend development improvement, after two weeks of testing, I can only say that there is still a big gap compared with Fabel. Kimi 3 (used through Cursor) also seems to perform better than 5.6 sol on some specific tasks.
Ultra mode should be completely removed. Please optimize your model or give me gpt6