Running Qwen 3.5-9B locally on an 8 GB 3070 Ti, I get around 41 tokens for everyday conversations and an inference speed of around 30 tokens. I think that’s still acceptable—not nearly as bad as people make it out to be, though it’s true that its intelligence has taken a hit.

Where do you think its greatest value lies?

In having no ethical constraints and being open to abuse for reverse engineering, cracking, unpacking, attacks, penetration testing...