Behind the scenes, the model is still reasoning with full thoughts, and then we have a summarization model which translates from full thoughts to summary, while preserving as much detail as possible.

→ View original post on X — @officiallogank,