Tencent Open Models: Model List, Licenses and Articles

September 26, 2026September 27, 2026

About Tencent

Tencent is a major Chinese technology company. It publishes 160 models on its tencent organization on Hugging Face, and its main language models are developed by the Tencent Hy Team (the GitHub organization is Tencent-Hunyuan).

It stands out for the breadth of what it releases, from large flagship models to small task-specific models for screen operation, document reading, translation and speech synthesis.

What It Releases

• Hy3 (July 2026) is an MoE model with 295B total and 21B active parameters. According to its model card, it scales up post-training with feedback from more than 50 products after the Hy3 Preview in late April, outperforms similar-size models and rivals flagship open models 2 to 5 times larger. The context is 256K tokens.
• Hy4 preview (August 2026) is a preview of the next flagship generation. Its backbone is an MoE with 770B total and 49B active parameters, and the model card says its architecture is inspired by DeepSeek and GLM, for example using Gated DSA, based on DeepSeek Sparse Attention.
• It also releases many small task-specific models, such as WeVisDoc for reading document images (2B and 4B; we have a family page), UI-Mate for screen-operating agents, Hy-MT2 for translation and AuK for speech synthesis.

Licenses

Hy3, Hy4 preview, WeVisDoc, UI-Mate and others are Apache-2.0, broadly usable including commercially. The AuK speech synthesis models are MIT. Some models, such as ContextPilot, use an other (custom) license, and some have no license field. Read each model’s license text before use. This summary is not legal advice.

Running Them Locally

• The flagship models do not fit on consumer GPUs. Hy3 activates 21B but needs all 295B parameters loaded. It has an official FP8 build aimed at data-center GPUs.
• Official GGUF builds are limited to models such as the Hy-MT2 translation models. Tencent itself also publishes GGUF files of the 1.8B quantized down to 1.25 and 2 bits. GGUF builds of Hy3 and Hy4 preview come from third parties.
• On consumer hardware, the small task-specific models are the realistic choice. WeVisDoc and UI-Mate-9B, for example, need little memory.

Sources: model cards of tencent/Hy3 and tencent/Hy4-preview; the model list of the tencent organization on Hugging Face (all as of 2026-09-26).

Our Coverage and Data

Local Model Watch has published 2 article(s) on Tencent’s own models, 0 on third-party fine-tunes and quantized builds based on them, and 0 from its official blog. The model list below is collected by code from Hugging Face every day. Part of our publisher index; model families link to their family pages.

At a Glance

Recent Models on Hugging Face

The publisher lists 160 models in total on Hugging Face. Below are up to 15 of the most recent generative models (text, image, video, audio and so on), the scope of this site, per organization. Classifiers, feature extractors, research components and repositories without a model card are omitted, and the publisher’s own quantized builds and helper variants are folded into the row of the original model. Parameter counts are left blank for repositories that pack weights into integer types, where Hugging Face’s count is not the real parameter count.

From the Hugging Face API. Last changed 2026-09-26 (JST). Collected by code.

Licenses of Its Own Models (as recorded in our articles)

The license can differ from model to model even within one publisher. This is not legal advice; check each model’s license text before use.

By Model Family

Articles on Its Own Models

Last updated 2026-09-26 (JST). The explanation at the top of this page was written with the help of AI from the primary sources it cites. The tables and lists under “Our Coverage and Data" are assembled by code from our article log and the Hugging Face API.