Follows directions very well
In some of my RP tests (with the AI playing a persona) Tiger did really great at following directions. For example one of the rules I have in the system message is that the assistant should never speak for anyone other than the assistant's character. Over the course of 10,000 tokens it only spoke for another character once or twice. But it wasn't any large sentences.
It also does well with ERP. Doesn't get anatomy wrong almost at all. I asked it to write a custom story about a person becoming someone else and it handled the transition really well. Most models start to fail with complexities like that.
My only wish for this model is for the context to be bigger like 32k. This is definitely one of my favorites. I guess I would also like some more flavour to the language like Hathor_Tahsin-L3-8B. But this model follows system messages much better than Hathor.
9/10 for me. (Disregarding 8k context)