when you distill a thinking LLM past its capacity it will default to overthinking because during training that was the only way for a chance at a reward on many tasks.
you can generally avoid this if you specialize it on a domain that is within its capacity.