some observations
This models is much more imaginative and capable of longer chains of logical hops, and feels astoundingly knowledgeable for its size. But it seems to confuse the one initiating the action with the one receiving it constantly. For example, if you say to your character "I turned into a werewolf", it'll be the character that wags its tail and gets fluffy claws.
It also has an interesting tendency to repeat the user's input verbatim. This seems to suggest some artefacts in the dataset or in the original nemo model's behavioral configs.but I can't be sure since I'm no expert.
But considering its size, it's an incredible piece of work overall, even though Snowpiercer v4 still feels like the most logically consistent and stable model in the 12-15B range for RP.