Hunyuan-A13B Instruct

Hunyuan-A13B is Tencent's open-source mixture-of-experts language model, released July 1, 2025, with 80 billion total parameters and only 13 billion active during inference. It was pretrained on a filtered 20-trillion-token corpus with enhanced STEM data curation, then fine-tuned with SFT and large-scale reinforcement learning.

Its signature is a dual-mode chain-of-thought framework: a fast-thinking mode handles routine queries with low latency while a slow-thinking mode engages in deeper multi-step reasoning for hard problems, adapting reasoning depth to task complexity. The model supports a 256K context window and native FP8 and Int4 quantization for efficient deployment.

Tencent reported competitive results on math and science reasoning, programming, and agent benchmarks such as BFCL-v3 and tau-bench, approaching much larger models. Weights are open under Tencent's Hunyuan license, whose custom terms should be reviewed before commercial use. The instruct variant is the chat-ready checkpoint for assistants and agents.