Tencent Releases Hy4-Preview, 770B Open-Weight MoE Model
Tencent has released Hy4-preview, a new open-source model with a MoE architecture: 770B total parameters, 49B active per token, and context up to 1 million tokens, with a focus on coding, agentic tasks, and productivity. It uses Gated DeepSeek Sparse Attention and IndexCache to lower the cost of long context and reuse sparse indexes across layers. Tencent positions Hy4 as a frontier-level open model with affordable deployment costs, licensed under Apache 2.0.
Related: Moonshot AI Releases Kimi K3 with 2.8 Trillion Open Weights, DeepSeek Launches V4 with 1 Million Token Context as Standard, Z.ai Releases GLM-5.3-Flash with Native Multimodal Support