Tencent Releases a 770-Billion-Parameter AI Model as a Free Download

Tencent released Hy4 preview on Aug. 28 and published the model's weights on Hugging Face, where the repository carries an Apache 2.0 license, terms that permit commercial use, modification and redistribution without a separate agreement.

The company's release describes Hy4 preview as a mixture-of-experts model, an architecture that routes each token through a small fraction of the network. It reports 770 billion total parameters, 49 billion active per token, and a context window above 1 million tokens. The model card on Hugging Face gives the same figures and adds the structure: 78 layers, of which 77 are mixture-of-experts layers holding 256 routed experts and one shared expert, with eight routed experts active per token.

Tencent says the model can be used through its WorkBuddy, CodeBuddy, Yuanbao and ima products, and reached by API through Tencent Cloud TokenHub and OpenRouter. Per the release, API pricing is $0.834 per million input tokens, $2.501 per million output tokens and $0.042 per million cached tokens. The company says Hy4 preview is free on WorkBuddy and CodeBuddy for two weeks after launch, and that free access to its previous model, Hy3, runs to Sept. 30.

The only quality comparison Tencent offers is its own. In a blind evaluation that the company says it ran internally with 163 experts across 203 engineering tasks, Hy4 preview averaged 2.99 out of 4.00, against 2.92 for GLM-5.3 and 2.94 for Kimi K3.

Tencent also says the model took part in optimizing its own training and inference software, proposing approaches, running experiments and feeding the results back, and that its work on the inference system raised end-to-end throughput 31.8% over the baseline.

Hy3 was released July 6, 2026, per Tencent's newsroom. The company says further Hy4 models will follow.