Adaptive reasoning LLMs based on the HiPO framework, featuring dynamic “think-on / think-off” control for efficient reasoning.