DS v4.1. Flash depreciating Performance

Hi, everyone! I have been using DS v4.1. Flash on dsh web, and it has been a powerhouse for me, and I was using it extensively since launch. But lately, I am noticing a huge dip in the quality and performance of this model. Sometimes, it just keep forgetting the context, or doesn't seem to follow the instructions provided, and sometimes, it just keep you in a never-ending loop of fixing a code / problem, where it pretends that it fixes everything, but it actually does not fix anything at all. However, GPT 5.6 did fix the same thing in just few minutes. I used to use Claude code, then shifted to GPT 5.6, but due to its 5-hours and weekly limit I had to switch to DS v4.1 flash as my daily driver especially for its speed and very cheap API. Now after the recent performance dip in v.4.1. flash, I'm thinking about moving back to GPT, but again GPT 6 Astra is almost unusable with Plus subscription and GPT 5.6 also started to eat a lot of 5 hrs+weekly usage on Plus, and I cannot afford to get their Pro subscription. So, I'm also thinking about GLM 5.3 API, but considering my usage, I dont think it's a good idea. I have also tried Qwen 3.8 27B locally which is great, but due to my limited 6GB VRam, its very slow. Has anyone else noticed the same performance in DS v.4.1. flash? What could be the reason? And what are the other alternatives we have now?

添加评论
点赞收藏
点踩分享查看原文
评论
?
参与讨论