Tencent Hunyuan Launches Hy4 Preview: 770B Sparse MoE Model Now on API

B.AI ·

Key Info

Tencent Hunyuan's Hy4 Preview, a 770B-parameter sparse MoE open-weights model that activates 49B parameters per token, is now live on the official API. It scored 2.99/4.00 in internal expert evaluations across 203 engineering tasks.

Highlights

  • 770B sparse MoE architecture with only 49B active parameters per token, offering efficient inference.
  • Open-weights model now available through official API access.
  • Strong internal evaluation result of 2.99/4.00 across 203 real-world engineering tasks.
Loading...