A compact reasoning model that thinks before it speaks — literally. It runs an internal chain-of-thought before producing answers, which helps it punch above its weight on logic and math problems for its size. The trade-off is latency: responses take longer than a straightforward 4B model would. Works well when accuracy matters more than speed.