That’s not surprising. I have been using Deepseek and it consistently produces excellent output given the right context howevwr. It depends on the task as it does have blindspots.