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

B.AI ·

Key Info

Hy4 Preview, the latest open-source 770B sparse MoE model from Tencent Hunyuan, is now available through the official API. It activates 49B parameters per token and delivered strong results in blind tests across 203 internal engineering tasks.

Highlights

  • Hy4 Preview is now fully accessible through official API channels.
  • Built on a 770B sparse MoE architecture, it activates only 49B parameters per token for efficient inference.
  • The model performed well in blind tests covering 203 real-world engineering tasks inside Tencent.
  • As an open-source release, it gives developers a new high-capacity option for AI agent and enterprise workloads.
Loading...