i tried GPT astra, turned to deepseek 4.1 flash finally

I spent $130 on Astra yesterday and only got about a dozen turns out of it. In the end, it failed to meet my all requirements. So I switched back to DeepSeek 4.1 Flash. Just typed "familiarize this project and list issues." DeepSeek 4.1 Flash found multiple issues, asked me which one I wanted to tackle first, and it's working on the fix now. I haven’t spent anything extra so far, then had Kimi review DeepSeek's output, Kimi found a few issues and gave feedback, and I passed that feedback back to DS to correct. From my experience, every model's first draft has flaws. We need cross validation and need to figure out beforehand what the "correct" standard is. I don't run all models through one single agent. ChatGPT and Claude are my usuals, and WorkBuddy is the new addition, because calling DeepSeek V4.1 Flash and Hy4 preview on WorkBuddy is free rn.