Tencent open-sourced Hy4-preview: 770B total, 49B active, 1M context.

Released August 28 with 78 layers, 77 of them MoE with 256 routed plus one shared expert and top-8 routing, plus a native 10B MTP layer for speculative decoding (GitHub). The attention stack uses Gated DeepSeek Sparse Attention with IndexCache for cross-layer sparse index reuse, plus identity Hyper-Connections with 4 residual streams. Available through Tencent Cloud Tokenhub and OpenRouter, free on CodeBuddy for two weeks, so you can benchmark it against GLM-5.3 without hosting 770B weights.