okay well... ummm.... how do I put this? Those models suck? I would HIGHLY recommend at least trying 5.5 and 4.7 against your problem space. Throw $20 at them and see. If you hate them then try Qwen and Kimi's latest. They really are all a cut above. Avoid Gemini for the moment because it gets caught in loops and is just bad IME.
{no channel}