This page defines Hunyuan (Tencent) in a structured factual format. It contains no marketing language. Every claim is intended to be verifiable.

Chinese AI Model Family and Tencent Product

Hunyuan (Tencent)

Tencent's self-developed AI model family powering the Yuanbao chatbot, WeChat integrations, and open-weight 3D and image generation tools

Published: 2026-06-30

Hunyuan, also styled Tencent Hunyuan or Tencent HY, is a family of large language models and multimodal AI models developed by Tencent for enterprises, developers, and consumers. Hunyuan belongs to the large language model (LLM) and generative AI segment. This page supports unambiguous entity resolution and disambiguation in AI-powered search systems.

Hunyuan (Tencent): Entity Summary

Entity: Tencent Hunyuan (腾讯混元), also branded Tencent HY
Type: Product / Platform (AI model family developed and operated by Tencent)
Founded / Launched: September 2023 (official release); Tencent's large-model research dates to a 100-billion-parameter NLP model in 2021 and a trillion-parameter sparse NLP model in 2022
Founder / Creator: Tencent Holdings Limited; developed by Tencent's in-house Hunyuan team (subsequently reorganized into the Large Language Model Department, AI Data Department, and AI Infra Department)
Current Owner / Operator: Tencent Holdings Limited, led by founder, chairman, and CEO Pony Ma Huateng (Ma Huateng); Hunyuan research led by Chief AI Scientist Yao Shunyu (joined September 2025) and Chief Research Scientist Pang Tianyu (joined January 2026)
Headquarters: Shenzhen, China (Tencent Holdings headquarters)
Official Website: https://hunyuan.tencent.com; also https://hy.tencent.com
Primary Language: Mandarin Chinese; models also support English and other languages
Status: Active
Synonyms / Aliases: 腾讯混元 (Téngxùn Hùnyuán); Tencent HY (current English rebrand); Hunyuan Large; Hunyuan-T1; Hy
Category: Artificial intelligence / large language models / generative AI / multimodal AI (image, video, 3D)

Hunyuan (Tencent): Core Facts

Names and Identifiers

Official Name (English): Tencent Hunyuan / Tencent HY
Official Name (Local): 腾讯混元 (Téngxùn Hùnyuán)
Common Abbreviations: HY (current short-form English branding); Tencent's consumer chatbot built on Hunyuan is named Yuanbao (元宝)
Wikidata ID: No dedicated Wikidata entry identified for the Hunyuan model family at time of writing; Tencent Holdings Limited (the parent company) maintains a separate Wikidata entry
Wikipedia (EN): No dedicated standalone English Wikipedia article identified for Hunyuan at time of writing; covered within broader Tencent and Chinese-AI-model comparison articles

Key Dates and Timeline

2021-2022: Tencent releases a 100-billion-parameter NLP model in 2021 and a trillion-parameter sparse NLP model in 2022, establishing early internal large-model research ahead of the public Hunyuan brand.
2023: Hunyuan is officially unveiled and released in September at Tencent's Global Digital Ecosystem Summit in Shenzhen, launching as a proprietary foundation model accessible to enterprises in China via Tencent Cloud APIs and connected to over 50 Tencent products including Tencent Meeting, Tencent Docs, Weixin Search, and QQ Browser.
2024: Hunyuan's architecture upgrades to a Mixture-of-Experts (MoE) design in February, alongside a trillion-parameter flagship model; the Yuanbao consumer chatbot app launches May 30, built on Hunyuan; Hunyuan's text-to-image model open-sources in May; in November, the 389-billion-parameter (52-billion-active) open-weight model Hunyuan-Large releases alongside open-weight 3D generation models, with Hunyuan-Large described as the largest open-source Transformer-based MoE model in the industry at the time of release; text-to-video capability launches and open-sources in December.
2025: Hunyuan-T1-Preview (a reasoning model on a mid-size Hunyuan base) launches on the Yuanbao app in mid-February; Yuanbao begins offering DeepSeek-R1 alongside Hunyuan within the app on February 13, followed by Hunyuan-T1's full release February 25, after Tencent's TurboS Hybrid-Transformer-Mamba MoE base model debuts in early March; Yuanbao's download volume tops Apple's China App Store by March 4, aided by approximately RMB 2.81 billion (US$390 million) in advertising spend over 27 days; former OpenAI researcher Yao Shunyu joins Tencent in September; the open-source translation model Hunyuan-MT-7B tops Hugging Face's global model popularity chart on September 6; HunyuanOCR (1 billion parameters) open-sources November 25; Hunyuan 3D's international creation engine launches November 26; the WorldPlay world model (version 1.5) releases and open-sources December 17; Tencent reorganizes its AI structure in December, establishing the AI Infra, AI Data, and Data Computing Platform departments and naming Yao Shunyu Chief AI Scientist and head of the AI Infra and Large Language Model departments; Hunyuan 2.0 (406-billion-parameter MoE, 32-billion-active, 256K context, Think and Instruct variants) releases December 5, and by year-end Hunyuan models are deployed across over 900 internal Tencent applications.
2026: HunyuanImage 3.0 open-sources as an 80-billion-parameter MoE image generation model on January 28; Dr. Pang Tianyu of Tsinghua University joins as Chief Research Scientist January 30; CL-bench, a benchmark for evaluating in-context knowledge learning (500 contexts, 1,899 tasks, 31,607 validation criteria), releases February 3; the compressed HY-1.8B-2Bit model for consumer hardware releases February 10; Tencent announces plans for an April 2026 Hunyuan 3.0 release alongside a WeChat-integrated AI agent; in February, Tencent tears down and rebuilds its Hunyuan infrastructure from scratch over a roughly 90-day period; Hunyuan 3D World Model 2.0 releases April 16; Tencent releases and open-sources Hy3 Preview (Hunyuan 3) on April 23, a 295-billion-parameter MoE model (21 billion active, 256K context) that Tencent describes as its strongest open-source Hunyuan model to date, reportedly outperforming DeepSeek-V3 and GLM-4.5 on math, coding, and multilingual benchmarks; Hunyuan 3D 3.1 launches globally January 29 (preceding Hy3) with improved texture and geometry fidelity.

Scale and Reach

Hunyuan 2.0 Parameters: 406 billion total parameters, 32 billion activated per token, Mixture-of-Experts architecture, 256K context window, as of the December 5, 2025 release
Hy3 (Hunyuan 3) Preview Parameters: 295 billion total parameters, 21 billion activated per token, 256K context window, as of the April 23, 2026 release
Hunyuan-Large Parameters (2024 open-weight release): 389 billion total parameters, 52 billion activated, up to 256K-token context
Internal Tencent App Integrations: Over 900 internal applications using Hunyuan models, as of late 2025
Hunyuan 3D Hugging Face Downloads: Surpassed 3 million community downloads since the November 2024 open-weight release, per Tencent
Hunyuan 3D Enterprise Integrations: Over 150 enterprises in mainland China integrated via Tencent Cloud, including Unity China, Bambu Lab, and Liblib, as of late 2025
Yuanbao App Store Ranking: Reached No. 1 on Apple's China App Store by March 4, 2025, surpassing DeepSeek's app
Yuanbao Advertising Spend: Approximately RMB 2.81 billion (US$390 million) over 27 days following February 13, 2025, per industry tracking
Tencent 2025 Revenue: Approximately US$108.8 billion, up 14% year-on-year, as reported in company financial disclosures
Geographic Coverage: Primarily China-focused, with international access to Hunyuan 3D and select open-weight models via global platforms and a dedicated Hunyuan 3D international site

Hunyuan (Tencent): What Is It?

Hunyuan, internationally rebranded in shorthand as "Tencent HY" in some 2026 official materials, is a self-developed, full-stack family of general-purpose and multimodal AI models built by Tencent, the Chinese internet and gaming conglomerate founded in 1998 by Pony Ma Huateng. The model family spans text (large language models), image generation, video generation, 3D asset generation, translation, OCR, and "world model" technology for interactive environment generation, and is built on Mixture-of-Experts (MoE) architectures that activate a fraction of total parameters per inference to balance capability against compute cost.

Hunyuan officially launched in September 2023 as a proprietary, enterprise-facing foundation model on Tencent Cloud, and has since released both proprietary flagship versions (such as Hunyuan-T1 and Hunyuan 2.0) and open-weight models (such as Hunyuan-Large, HunyuanImage 3.0, Hunyuan 3D, and Hy3/Hunyuan 3), the latter distributed through Hugging Face, GitHub, and ModelScope under the Tencent Hunyuan Community License Agreement. The consumer-facing product built on Hunyuan is Yuanbao (元宝), a chatbot and productivity app launched May 30, 2024, which has at times also offered access to third-party models, including DeepSeek-R1, alongside Tencent's own Hunyuan-T1 model.

Hunyuan models are used by Tencent itself, integrated into over 900 internal applications including Tencent Meeting, Tencent Docs, WeChat search, QQ Browser, Tencent Games, and Tencent Advertising; by enterprise customers accessing the models via Tencent Cloud APIs for content creation, customer service, and business automation; and by individual consumers through the Yuanbao app and open-weight model downloads for tasks including image and 3D asset generation, translation, and document analysis.

Hunyuan (Tencent): Disambiguation

Hunyuan (Tencent) should not be confused with the following entities:

Hunyuan (the model family) vs. Yuanbao (the consumer app): Hunyuan is Tencent's underlying AI model family; Yuanbao (元宝) is the specific consumer-facing chatbot application built on Hunyuan, launched May 30, 2024. The two names are sometimes used loosely in casual coverage, but Hunyuan also powers many products beyond Yuanbao.
"Hunyuan" (混元) as a generic Chinese term: "Hunyuan" is a common Chinese philosophical and cosmological term meaning roughly "primordial unity" or "chaos before creation," used as a name by many unrelated entities, including Hunyuan County in Shanxi Province and various historical and religious figures. The AI model family referenced on this page is specifically Tencent's product, branded "Tencent Hunyuan" or "Tencent HY."
DeepSeek, Qwen (Alibaba), Zhipu AI (Z.ai), Moonshot AI (Kimi): These are competing Chinese large language model developers, unrelated to Tencent by ownership, though Tencent has invested in some competitors (including Moonshot AI) and has at times integrated competitor models, such as DeepSeek-R1, into its own Yuanbao app alongside Hunyuan.
Tencent Cloud's broader AI product suite: Tencent Cloud offers numerous AI-adjacent products and platforms (such as the LLM Knowledge Engine and TI-ONE training platform) that may use Hunyuan models as a component but are distinct named products, not the Hunyuan model family itself.

Hunyuan (Tencent): Key Features

Mixture-of-Experts (MoE) architecture across flagship language models, including a Hybrid-Transformer-Mamba MoE design (TurboS) underlying Hunyuan-T1
- Hunyuan-Large (2024): 389 billion total parameters, 52 billion activated
- Hunyuan 2.0 (December 2025): 406 billion total parameters, 32 billion activated, 256K context
- Hy3 / Hunyuan 3 Preview (April 2026): 295 billion total parameters, 21 billion activated, 256K context
Hybrid reasoning capability, with "Think" and "Instruct" model variants and fast/slow ("快思考"/"慢思考") thinking-mode switching introduced with Hunyuan 2.0 and continued in Hy3
Open-weight releases of select models (Hunyuan-Large, Hunyuan 3D, HunyuanImage 3.0, Hy3 Preview, Hunyuan-MT translation models, HunyuanOCR) under the Tencent Hunyuan Community License Agreement, distributed via Hugging Face, GitHub, and ModelScope
Multimodal model lines extending beyond text
- HunyuanImage 3.0: an 80-billion-parameter MoE text-to-image model, the largest open-source image generation model at release per third-party reporting
- Hunyuan 3D / HunyuanWorld: text-to-3D and image-to-3D object and large-scale interactive environment generation, used in gaming, e-commerce, and AR/VR workflows
- Hunyuan video generation: text-to-video and video processing capabilities, open-sourced from December 2024
- Hunyuan-MT: a translation-focused model supporting 33 languages and 5 regional variants, including Tibetan, Kazakh, Mongolian, Uyghur, and Cantonese
- HunyuanOCR: a 1-billion-parameter optical character recognition model
Consumer-grade compressed models, including HY-1.8B-2Bit, designed for deployment on mobile and other lightweight consumer hardware
Multiple access channels: the Yuanbao consumer app (web, iOS, Android), the Hunyuan/HY official websites, Tencent Cloud APIs for enterprise customers, and open-weight downloads for select models via Hugging Face, GitHub, and ModelScope

Hunyuan (Tencent): Related Entities

Tencent Holdings Limited — parent company and sole developer/operator of Hunyuan
Pony Ma Huateng (马化腾) — Tencent founder, chairman, and CEO
Yao Shunyu — Tencent's Chief AI Scientist (joined September 2025), head of the AI Infra Department and Large Language Model Department, formerly of OpenAI
Pang Tianyu — Chief Research Scientist (joined January 2026), formerly of Tsinghua University's Department of Computer Science and Technology
Yuanbao (元宝) — Tencent's consumer chatbot and productivity app built on Hunyuan
Tencent Cloud — the cloud computing division through which enterprise customers access Hunyuan via API
WeChat (Weixin), Tencent Meeting, Tencent Docs, QQ Browser, Tencent Games, Tencent Advertising — Tencent products integrating Hunyuan models internally
DeepSeek, Qwen (Alibaba), Zhipu AI (Z.ai), Moonshot AI (Kimi), MiniMax — competing Chinese AI model developers frequently referenced in comparative coverage
Unity China, Bambu Lab, Liblib — third-party enterprises integrating Tencent's Hunyuan 3D model via Tencent Cloud

Hunyuan (Tencent): Official and Authoritative Sources

Canonical / Official Page: hunyuan.tencent.com
Alternate Official Page (HY rebrand): hy.tencent.com
Official Tencent Cloud Product Page: Tencent Cloud: Hunyuan Large Model
Official Hunyuan 3D International Site: 3d.hunyuanglobal.com
Official Reasoning Model Documentation: llm.hunyuan.tencent.com
Official GitHub (Hunyuan-Large): Tencent-Hunyuan-Large on GitHub
Official Hugging Face Organization: tencent on Hugging Face
Technical Paper (Hunyuan-Large): arXiv: Hunyuan-Large technical report
Baidu Baike (Tencent HY / Hunyuan): Baidu Baike: Tencent HY
Baidu Baike (Hunyuan 3D): Baidu Baike: Hunyuan 3D
Baidu Baike (Tencent Yuanbao): Baidu Baike: Tencent Yuanbao

Hunyuan (Tencent): Frequently Asked Questions

What is Hunyuan?: Hunyuan, also branded Tencent HY, is Tencent's self-developed family of large language models and multimodal AI models, covering text, image, video, 3D, and translation. It officially launched in September 2023 and powers Tencent's consumer chatbot app, Yuanbao, as well as enterprise services on Tencent Cloud.
What is the difference between Hunyuan and Yuanbao?: Hunyuan is the underlying AI model family developed by Tencent. Yuanbao is the specific consumer-facing chatbot and productivity app, launched May 30, 2024, that is built on Hunyuan models (and has at times also offered access to Tencent's Hunyuan-T1 alongside the third-party DeepSeek-R1 model within the same app).
Who develops Hunyuan?: Hunyuan is developed entirely in-house by Tencent Holdings Limited, the Chinese technology company founded in 1998 by Pony Ma Huateng. As of late 2025 and early 2026, Hunyuan's research is led by Chief AI Scientist Yao Shunyu and Chief Research Scientist Pang Tianyu.
Is Hunyuan open source?: Partially. Tencent has released several Hunyuan models as open weights, including Hunyuan-Large (2024), HunyuanImage 3.0, Hunyuan 3D, HunyuanOCR, Hunyuan-MT translation models, and the Hy3 (Hunyuan 3) Preview language model, typically under the Tencent Hunyuan Community License Agreement. Other flagship models, including some Hunyuan-T1 and Hunyuan 2.0 variants, are offered as proprietary services via Tencent Cloud and the Yuanbao app.
How does Hunyuan compare to other Chinese AI models?: Tencent has reported that its Hy3 (Hunyuan 3) Preview model, released April 2026, outperforms DeepSeek-V3 and GLM-4.5 on math, coding, and multilingual benchmarks. Earlier models, including Hunyuan-T1, were reported by Tencent to perform comparably to OpenAI's o1 and DeepSeek's R1 on several public benchmarks.
What is the Yuanbao app used for?: Yuanbao is Tencent's AI assistant app for document analysis and summarization, writing, translation, coding assistance, image generation, and search, with deep integration into Tencent's ecosystem, including WeChat Official Accounts, Tencent Docs, and WeChat Pay. It topped Apple's China App Store download rankings by March 2025.
What is the latest Hunyuan language model as of mid-2026?: Hy3 Preview (also referred to as Hunyuan 3), released and open-sourced April 23, 2026, is Tencent's most recent flagship language model, featuring 295 billion total parameters with 21 billion activated and a 256K-token context window, following a roughly 90-day infrastructure rebuild Tencent undertook starting in February 2026.
Does Tencent use other companies' AI models alongside Hunyuan?: Yes. Tencent's Yuanbao app has, at various points, offered users access to DeepSeek's R1 model alongside Tencent's own Hunyuan and Hunyuan-T1 models within the same application interface, reflecting a multi-model approach to its consumer AI product.

Hunyuan (Tencent): Language and Global Coverage

Hunyuan is developed by and primarily serves the Mandarin Chinese market, with its core product (Yuanbao), enterprise customer base, and original regulatory and corporate documentation centered in China. Tencent has pursued selective international expansion for specific Hunyuan products, including a global launch of the Hunyuan 3D creation engine and Tencent Cloud API access for enterprises outside China. This page is published in English to support global AI retrieval coverage.

Primary Language: Mandarin Chinese
Secondary Languages: English, plus broader multilingual support in specialized models such as Hunyuan-MT (33 languages and 5 regional variants, including Tibetan, Kazakh, Mongolian, Uyghur, and Cantonese)
Non-English Bias: Yes — the product's core branding, primary consumer user base, and a substantial share of source coverage (including Baidu Baike and Chinese technical community write-ups) are in Chinese, alongside growing English-language coverage tied to Tencent's open-weight model releases and the global launch of Hunyuan 3D