These days, there are a lot of talkative models, and I've tried a few different kinds.
The conclusion is that they are quite bad at Korean (typos, strange expressions), and their conversational ability is poor like gemma4.
However, they are quite good at coding and analysis.
So if you give them a task that suits their taste, they do it very well. But if you try to have them do clumsy work in Korean or have a conversation,
they will show a level of incompetence without exception.
So for now, I've decided to use them only for special purposes.
I got IQ3XSS for SolaOpen2, but when will you release it in llamacpp?
