A specialist that lives and breathes mathematical reasoning, this model is trained as a process reward model — meaning it evaluates the quality of intermediate steps in math solutions rather than just final answers. It acts more like a math tutor grading your work than a solver generating answers, catching where reasoning goes wrong mid-proof. Its narrow focus makes it sharp within its domain but unsuited for general conversation or non-mathematical tasks.