Kwaipilot/HiPO-8B
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Sep 26, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Kwaipilot/HiPO-8B is an 8 billion parameter language model developed by Kwaipilot, featuring a 32768 token context length. It utilizes Hybrid Policy Optimization (HiPO), a novel RL framework, to dynamically decide between 'Think-on' (reasoning) and 'Think-off' (direct answer) modes. This model is optimized for balancing reasoning accuracy with efficiency, achieving significant improvements in both metrics by reducing token length and thinking rate.

Loading preview...