Tencent releases Hy4 Preview, a 770B-parameter open-weight LLM with 1M token context

Tencent released Hy4 Preview on August 29, 2026, according to a report by Simon Willison. The model is an open-weight, text-only large language model, with no vision capability. Willison’s summary lists 770 billion total parameters, with 49 billion active parameters, and a context window of one million tokens. The weights are hosted on Hugging Face, occupying 1.56 terabytes.

Willison notes this represents a substantial size increase over Tencent’s previous release, Hy3, published in July. That earlier model had 295 billion total parameters, 21 billion active parameters, and a 256,000 token context window.

The Hy4 Preview announcement itself was published at hy.tencent.ai under the title “Introducing Hy4 Preview.” Willison’s post links both to Tencent’s research page and to the Hugging Face repository hosting the model weights.

The release continues a pattern of large open-weight model releases from Chinese technology companies through 2026, with Tencent’s own output growing markedly between its July and August releases in both total and active parameter counts, as well as context length. No further architectural or benchmark detail beyond these figures appears in the cited summary.