Qwen 3.6 27B
Hi @Gryphe , thanks for posting another model—I'm a big fan of your fine-tunes, especially Codex.
Is there any chance you’d take a look at Qwen 3.6 27B? This model is a smart beast with a very efficient KV Cache, but so far no one has managed to create a really good fine tune designed for roleplay.
It would be great to see a new version of Codex (or Pantheon) based on the new Qwen.
The main challenge with all these new and shiny models is that they seem to produce their best work when there's reasoning included in their training, and that's a bit I still need to figure out for roleplay without the result feeling repetitive. (Reasoning and repetition seem to go hand in hand, or so it feels)
As you're probably aware I don't release new models frequently but I'll definitely look into doing a dense model next either way.