The new version features increased thinking depth, evidenced by a higher token usage per question in complex tasks, leading to improved accuracy, reduced hallucination rates, enhanced function calling support, and better "vibe coding" experience.