Key Info

vLLM announced Day-0 support for Tencent's Hy4, a natively multimodal model that processes text, images, and video within a single model.

Highlights

  • Hy4 is supported in vLLM from launch day
  • The model handles text, images, and video in one unified model
  • Use cases cited include reading visual documents, comparing images, and reasoning over video clips alongside text
  • Hy4 is a 770B total / 49B active MoE with 1M context, released with open weights