Subscribe

Hy4 preview

770B-total, 49B-active open MoE with 1M-token context, Gated DeepSeek Sparse Attention and hyper-connections, under Apache-2.0.

A step up from Hy3 in size, context and data, with 78 layers, 256 routed experts and a 10B MTP layer. Per its model card the architecture is inspired by DeepSeek and GLM. Tencent flags over-long reasoning as a known flaw. Hy4 preview topped OpenRouter token volume in early September (Wikipedia).

Date
Friday 28 August 2026
Lab
Tencent
Kind
open-weights
Access
open weights

Figures

MeasureValueMeasured by
Total / active parameters770B / 49Bcompany
Context length1Mcompany
OpenRouter weekly tokens (week to 2026-09-06)14.7T
ranked first on OpenRouter, per Wikipedia citing OpenRouter
independent

Released 2026-08-28 according to Wikipedia and the OpenRouter listing, with the Hugging Face repo created 2026-08-27 UTC. The model card gives its benchmark results only as images. Bloomberg reported that Tencent claims Hy4 beats Z.ai and Moonshot models.

Sources

  1. huggingface.co/tencent/Hy4-preview
  2. en.wikipedia.org/wiki/Tencent_Hy

This record was checked against its sources on 6 October 2026. How we check

Related