Hunyuan-A13B Technical Report

Tencent introduces Hunyuan-A13B, an open-source Mixture-of-Experts model with 80 billion total parameters and 13 billion active during inference. It utilizes a 20T-token corpus and a dual-mode Chain-of-Thought framework for optimized reasoning. Simultaneously, the Pistis model family, featuring 27B and 9B parameter multimodal models based on Qwen3.6 and Qwen3.5, is launched. Pistis employs a scalable post-training framework using Interleaved Distillation and Reinforcement Learning to integrate on-policy distillation and reinforcement learning within one training loop.