LaunchAI ModelsJuly 2, 2026
Tencent releases Hy3, a 295B MoE model with 21B active parameters

Hy3 is a 295B-parameter MoE model that activates only 21B parameters per token, licensed under Apache 2.0 with 256K context. It reportedly beats GLM-5.1 in blind tests while running fewer active params, using plain GQA architecture for efficiency.
5 sources
More stories today
- Alibaba reportedly tests standalone Qwen Office product
- Google shares Gemini usage data: multimodal AI useful for manual labor
- South Korea outlines AI future with NVIDIA at AI Summit
- PicoAgents framework for multi-agent systems released
- HuggingHack local HuggingFace tool moves to GitHub