Tencent is the Chinese internet conglomerate behind WeChat, and its Hunyuan program is one of the most consequential model efforts among Chinese AI labs by distribution alone: WeChat's billion-plus users and the Yuanbao assistant give Tencent consumer reach no dedicated lab can match.1

Stated mission and record

Tencent states its corporate vision and mission as "Value for Users, Tech for Good."4 The expanded formulation on its corporate site reads: "User value is our guiding principle, we strive to incorporate social responsibility into our products and services; promote technology innovation and cultural vitality; help industries digitally upgrade; collaborate for the sustainable development of society."4 The company pairs this with four stated values (integrity, proactivity, collaboration, creativity), but its corporate site presents no AI-specific charter comparable to the founding documents of dedicated labs; the AI work sits inside this general corporate framework.4

The record shows both consistencies and tensions with that framing, as of July 2026. Consistent with it: the Hunyuan program moved from a closed enterprise line to broad open-weight releases, with HunyuanVideo and Hunyuan3D published openly, Hunyuan 3.0 shipped under Apache 2.0 per July 2026 reporting, and Tencent's verified Hugging Face organization hosting roughly 140 models.5 The early 2025 integration of DeepSeek-R1 into WeChat search and Yuanbao placed a rival's open model in front of Tencent's own users, a choice contemporaneous coverage read as user-value pragmatism.3 In tension: the flagship Hunyuan text models stayed closed from the program's 2023 launch until Hunyuan-Large in November 2024, and the mission language sets no specific commitments on model release, safety evaluation, or transparency, so observers differ on how to measure the AI program against it.

Model program

Hunyuan launched in 2023 as a closed enterprise line, then turned decisively open: Hunyuan-Large (November 2024) released a 389-billion-parameter MoE with 52 billion active, then the largest open MoE by total parameters, with a documented technical report.1 HunyuanVideo (December 2024) became the leading open video-generation model of its moment, and Hunyuan3D opened asset generation; per July 2026 reporting, Hunyuan 3.0 ("Hy3," 295B total, 21B active, 256K context) shipped under Apache 2.0, continuing the open cadence.2

Strategy

Tencent's distinguishing move was pragmatism over pride: in early 2025 it integrated DeepSeek-R1 directly into WeChat search and Yuanbao alongside its own models, validating open-weight adoption at maximum scale and pressuring rivals to justify closed stacks. Its games, advertising, and cloud businesses absorb Hunyuan capabilities internally, and its capital ties reach across the landscape (an early backer of Moonshot AI among others).3

Position

Hunyuan competes with the Qwen and DeepSeek lines on the open leaderboards and with Baidu's ERNIE for enterprise deals; its video and 3D models are the program's clearest world-leading niches, benchmarked against Western rivals from Google DeepMind's Veo line to the open diffusion ecosystem of Stability AI descendants. Capability claims carry developer attribution per this wiki's rules.23

See also

References


  1. Tencent Hunyuan Team, "Hunyuan-Large: An Open-Source MoE Model with 52 Billion Activated Parameters," arXiv:2411.02265, November 2024. ↩↩

  2. Tencent Hunyuan release materials 2024-2026; Hunyuan 3.0 (Hy3) coverage in Chinese-model roundups, July 2026. ↩↩

  3. Coverage of DeepSeek integration into WeChat/Yuanbao, February 2025 (Reuters). ↩↩↩

  4. Tencent, "About Tencent" (Vision & Mission and Values sections), official website, as of July 2026. ↩↩↩

  5. Tencent, organization page, Hugging Face, as of July 2026. ↩