What Tencent AI Lab does

Tencent AI Lab is the corporate-level AI research and application division of Tencent Holdings Ltd (HKEX: 0700), founded in 2016 and headquartered in Shenzhen, China. The lab develops large language models, multimodal AI systems, and AI for life sciences, serving as Tencent's central R&D engine for foundation model technology. Its portfolio spans the Hunyuan platform, including Hy3 (a 295B total / 21B active parameter MoE LLM with hybrid fast-and-slow thinking architecture), Covo-Audio (a 7B end-to-end Large Audio Language Model), WorldMirror (Hunyuan World Model v1.1 for single-GPU real-time 3D scene generation), HunyuanVideo (open-source text-to-video), CogVLM (vision-language model), and the Ontology Reinforcement Iteration (ORI) protein design framework operated through the Tencent AI for Life Sciences Lab.

The lab runs a dual go-to-market: open-source distribution under Apache 2.0 through global ML platforms (Hugging Face, ModelScope, OpenRouter, GitHub) targeting researchers and developers, and OEM-style integration into Tencent's internal product ecosystem (WorkBuddy for enterprise productivity, Yuanbao for consumer AI, Marvis for enterprise AI, and ima for knowledge management). The latter provides distribution to hundreds of millions of users and has driven a 20x increase in daily token consumption and 6x growth in WorkBuddy Hy3 users since the April 2026 preview. The lab does not monetize directly; it is a non-revenue-generating R&D division whose value capture occurs indirectly via Tencent's platform and product revenue.

Strategic partnerships include a co-development and wet-lab validation agreement with Sino Biological (May 2026), published in Nature Communications, and ecosystem sponsorships of the OpenClaw AI community and China's Super Computing Network, where the lab supports aggressive token pricing (0.1 yuan per million tokens). Customer concentration is effectively total within Tencent's internal products, with external adoption measured through open-source downloads and developer engagement rather than contracted revenue.

Name

Tencent AI Lab

Legal name

Tencent AI Lab (Tencent Holdings Ltd)

Website

https://ai.tencent.com

Company type

Private

Founded year

2016

Operating status

Operating

Short description

Tencent AI Lab is the AI research division of Tencent Holdings, developing large language models (Hy3), multimodal AI (Covo-Audio, WorldMirror, HunyuanVideo), and AI-driven protein design. It open-sources models under Apache 2.0 and integrates them across Tencent's WorkBuddy, Yuanbao, Marvis, and ima products serving hundreds of millions of users.

Ownership category

akta.pro rank

Product category

AI Foundation Models

NAICS

Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (5182), Computer Systems Design and Related Services (5415), Other Computer Related Services (541519)

SIC

Services-Prepackaged Software (7372), Services-Computer Integrated Systems Design (7373)

akta.pro primary industry

Open-Source Model Ecosystems & Model Marketplaces (HDAAACAM)

akta.pro secondary industries

Enterprise Foundation Model Integration & APIs (Connectors, Governance, Deployment) (HDAAACAO), Model Development & Training Platforms (AutoML, Notebooks, Feature Stores) (HDAEANAB), Model Hosting, Serving & Inference Platforms (HDAAACAB)

Keywords

• Large language models
• Multimodal AI models
• AI research lab
• Open source AI
• AI protein design

Headquarters

HQ city

Shenzhen

HQ country

China

HQ region

Asia

Offices2 records

Markets served

GTM type

B2B

Offering type

Software

Cost components

Personnel, Technology or R&D, Infrastructure, Marketing or Sales

Revenue model

1. Open Source Model Distribution: Tencent AI Lab releases models under Apache 2.0 open-source license, generating value through developer adoption, ecosystem building, and indirect monetization through Tencent's integrated product ecosystem.
2. Internal Product Integration: Models integrated into Tencent's internal products (WorkBuddy, Yuanbao, Marvis, ima) drive user engagement and retention across Tencent's platform ecosystem, contributing to Tencent's overall revenue.

Pricing tiers

Go-to-market motion2 records

Distribution channels4 records

Marketing channels5 records

Core offering

Tencent AI Lab is a corporate-level AI research division that develops and open-sources large language models and multimodal AI systems under the Hunyuan brand. Its flagship Hy3 is a 295B-total-parameter Mixture-of-Experts LLM with a hybrid fast-and-slow thinking architecture, complemented by WorldMirror (3D generation), Covo-Audio (end-to-end speech-language model), HunyuanVideo (text-to-video), and CogVLM (vision-language). The models are released under Apache 2.0 and embedded into Tencent's enterprise and consumer products.

Product overview

Tencent AI Lab operates as a multi-product AI research and deployment organization built around the Hunyuan platform. Its flagship offering is Hy3, a 295B-parameter Mixture-of-Experts LLM with hybrid fast-and-slow thinking architecture released under Apache 2.0 in July 2026. Complementary core products include HunyuanVideo (open-source text-to-video model), WorldMirror (v1.1 of the Hunyuan World Model for 3D scene creation from multimodal inputs), Covo-Audio (7B-parameter end-to-end speech language model), and CogVLM (open-source multimodal vision-language model). These models power integrated Tencent consumer and enterprise products including WorkBuddy, Yuanbao, Marvis, and ima. A dedicated Tencent AI for Life Sciences Lab applies AI to protein design using the Ontology Reinforcement Iteration framework. The Hunyuan platform completed its full model-development loop in under six months, with Hy3 available via Hugging Face, ModelScope, and OpenRouter.

Differentiator

Problem solved

Functional benefit

Brands

• Hy3 (Hunyuan 3): 295B total parameter Mixture of Experts (MoE) model with hybrid fast-and-slow thinking architecture and 21B active parameters
• Covo-Audio
• WorldMirror
• HunyuanVideo
• Hunyuan

Products and services

• Hy3 (Hunyuan 3) Hy3 is Tencent's flagship large language model with a hybrid fast-and-slow thinking architecture featuring 295 billion total parameters and 21 billion active parameters, delivering performance comparable to flagship models at significantly lower computational cost. Available globally under Apache 2.0 open-source license for researchers, developers, and enterprise users.
• WorldMirror (Hunyuan World Model v1.1) WorldMirror is a 3D scene generation model supporting multi-view, video, camera, and depth inputs for end-to-end 3D scene creation. Runs in real time on a single GPU and is fully open source on GitHub with a Hugging Face Space for online testing.
• Covo-Audio Covo-Audio is a 7B-parameter end-to-end Large Audio Language Model that directly processes continuous audio inputs and generates audio outputs within a single unified architecture, eliminating the need for traditional cascaded ASR-LLM-TTS pipelines. Targets developers of voice-first AI applications.
• HunyuanVideo HunyuanVideo is Tencent's open-source text-to-video generation model, cited among top open-source video AI models alongside Mochi and Wan2.2 and approaching the quality of proprietary systems like Kling and Sora.
• CogVLM CogVLM is an open-source multimodal AI model that processes both text and images for applications including document understanding, visual question answering, and autonomous vehicles.
• Tencent Hunyuan (foundation model platform) Tencent Hunyuan is the overarching AI foundation model platform encompassing the Hunyuan LLM family, HunyuanVideo, the Hunyuan World Model, and related AI capabilities. It serves as the umbrella platform that completed its full model-development loop in under six months.

Quantifiable outcome

• 20x increase in daily token consumption across Tencent products since Hy3 preview
• +5 more outcomes

Named customers4 records

Segments4 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability10 records

Feature6 records

Partnerships

Three partnerships are on record, tiered core and minor.

• Sino BiologicalcoreStrategic or Co-development Partner · 28 May 2026Sino Biological collaborated with Tencent AI for Life Sciences Lab on a study published in Nature Communications. Sino Biological's cell-free protein synthesis workflow, including the XPressMAX™ kit, was used to enable rapid validation of AI-designed proteins created using the ORI framework. This collaboration bridges AI computational protein design with experimental wet-lab validation.
• OpenClaw AI CommunityminorGTM or Marketing Partner · 16 March 2026Tencent Cloud and Tencent AI became sponsors of the OpenClaw AI community following resolution of a data scraping dispute. Tencent Cloud provides lightweight application servers for one-click deployment and a free installation program across 17 cities in China, along with enterprise-grade security solutions.
• China's Super Computing NetworkminorGTM or Marketing Partner · 12 March 2026China's Super Computing Network launched an OpenClaw promotion with 10 million free tokens per user and pricing as low as 0.1 yuan per million tokens. Tencent Cloud became a sponsor of this promotion, with Tencent AI Lab indirectly benefiting through the partnership.

Scale indicators7 records

Recent moves6 records

Expansion highlights5 records

Direct peers

• DeepSeek: DeepSeek is a Chinese AI company developing competitive open-weight large language models (e.g., DeepSeek-V3, R1) released under permissive licenses for global developer adoption. It is the closest direct competitor to Tencent AI Lab's Hy3 in the Chinese open-source LLM category, targeting the same developer and enterprise audiences with comparable MoE-style architectures.
• Alibaba Qwen (Alibaba Cloud): Alibaba's Qwen team releases one of the most-downloaded families of open-weight LLMs globally (Qwen2/3, Qwen-VL, Qwen-Audio), covering text, vision, and audio modalities. It competes head-to-head with Tencent AI Lab's Hy3, Covo-Audio, and CogVLM, and similarly distributes through Hugging Face and ModelScope.
• Zhipu AI: Zhipu AI is a Chinese foundation-model company behind the GLM/ChatGLM family of open-weight LLMs and the agent-focused AutoGLM products. It competes with Tencent AI Lab in open-source Chinese LLMs and increasingly in agentic AI capabilities embedded in enterprise products.
• Baichuan Inc. Baichuan develops Baichuan-series open-source large language models aimed at Chinese enterprises and developers. It is a direct Chinese open-weight LLM peer to Tencent AI Lab's Hy3 with similar target market, licensing posture, and distribution channels.
• ByteDance Doubao Team: ByteDance's Doubao team develops large multimodal foundation models (text, image, video) and integrates them across ByteDance's consumer products (Doubao, TikTok/ByteDance apps). It mirrors Tencent AI Lab's strategy of combining internal model development with deep integration into a major Chinese consumer platform.
• Mistral AI: Mistral AI is a European open-weight foundation-model developer releasing competitive LLMs (Mistral, Mixtral) under Apache 2.0-style licenses. It overlaps with Tencent AI Lab on open-weight distribution, multimodal ambitions, and serving as a lower-cost alternative to US frontier labs.

Broad incumbents

• Meta AI (Llama): Meta AI's Llama family is the dominant open-weight LLM globally, distributed under a custom license via Hugging Face and integrated across Meta's products. It is a broader incumbent peer to Tencent AI Lab's Hy3 in the open-source LLM category, with vastly larger distribution and developer mindshare.
• OpenAI: OpenAI is the leading proprietary foundation-model developer (GPT-4o, o-series), offering closed-weight models via API and ChatGPT. It sets the frontier benchmark Tencent AI Lab's Hy3 aims to match at lower cost, and competes for the same global enterprise AI budgets.
• Anthropic: Anthropic develops the Claude family of frontier LLMs focused on reasoning, safety, and enterprise deployment. It competes with Tencent AI Lab for global enterprise AI mindshare and benchmarks, even though it does not release open weights.
• Google DeepMind (Gemini): Google DeepMind develops the Gemini family of multimodal foundation models (text, image, audio, video) integrated across Google products. It is a broader incumbent peer to Tencent AI Lab's full-stack multimodal strategy (Hy3, Covo-Audio, WorldMirror, HunyuanVideo).

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights7 records

Customer concentration

• LinkedInlinkedin.com/company/tencentglobal

Revenue estimate

Valuation estimate

Number of profiles

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Tencent AI Lab