Deepseek got better for Creative Writing!!!

For many of you that know, the first flash version had a massive downgrade when writing, whether it's creative or roleplay. I remember wrestling with it for minutes trying to get it to think and follow a style but it doesn't reciprocate anything at all. I found out it's because it's been heavily optimized for code, thus cutting down its writing capabilities, so for the time being I used GLM/Qwen, which weren't impressive but... better, at least they think. For this model, it still suffered from the same prose/dialogue issue but I came to see that it thinks when you point it out, or improve it's prose when you instruct it. I found that it follows your suggestions better if you're not specific, like straight up calling it's AI wording choice - go around. Anyway!!! Excited for Deepseek V4.1 Pro