Tencent HY poaches former OpenAI scientist to lead multimodal AI efforts
Tencent is reportedly hiring former OpenAI researcher Yonglong Tian to lead its Hunyuan multimodal model team, with the move expected to report to chief AI scientist Yao Shunyu. The hiring highlights Tencent's focus on concentrating talent for vision-language research, though the report currently remains thinly corroborated.
Key Takeaways
- Yonglong Tian, former OpenAI research scientist and MIT PhD, joins Tencent to head the Hunyuan multimodal model team.
- Tian will report directly to Yao Shunyu, Tencent’s Chief AI Scientist, who was recruited from OpenAI in late 2025.
- The hiring focuses on strengthening Tencent's vision-language research, building on Tian's expertise in supervised contrastive learning and fluid autoregressive models.
- The move follows the recent July 2026 launch of Hunyuan Hy3, a reasoning model utilizing personal assistant and coding integrations.
Why It Matters
The recruitment of Tian signals an aggressive effort by Tencent to close the gap with regional competitors like ByteDance and Alibaba in native multimodal capabilities. By centralizing talent with direct experience in OpenAI’s training workflows, Tencent aims to accelerate the integration of vision-language models into its gaming and social ecosystems. This talent shift reflects a broader trend of high-level AI researchers moving from U.S. labs to Chinese tech giants, potentially accelerating the development of sovereign foundational models in Asia. Watch for upcoming Hunyuan model cards or technical reports to see if Tian’s research in continuous token image generation is implemented in near-term 3D or video synthesis updates.
Additional Context
Tencent’s recruitment of Yonglong Tian follows the high-profile appointment of Yao Shunyu in December 2025, who was reportedly hired with a compensation package exceeding 100 million yuan, per Caixin Global (January 2026). At just 28, Yao was placed in charge of Tencent’s newly established AI infrastructure and large language model departments, reporting directly to President Martin Lau. This structural shift highlights Tencent's move away from fragmented incremental AI upgrades toward a unified architecture under the Hunyuan brand, which now powers over 900 internal business scenarios including the WeChat-linked 'Yuanbao' assistant. Technically, the Hunyuan ecosystem has seen rapid expansion throughout 2026. In May 2026, Tencent released 'HY-World 2.0,' a multimodal world model designed for 3D reconstruction and simulation, which the company positioned as a competitor to Google’s Genie series, per internal technical reports (May 2026). More recently, in early July 2026, Tencent launched its Hunyuan Hy3 model, which adopted a Mixture-of-Experts (MoE) architecture with 295 billion parameters, demonstrating the lab's capacity to ship high-scale reasoning models according to TechNode (July 2026). Tian brings a specific technical pedigree in representation learning and computer vision. His academic record includes foundational work on 'Supervised Contrastive Learning' and more recent research into scaling autoregressive models using continuous tokens rather than discrete vector quantization, per Pandaily (July 2026). This expertise is particularly relevant to Tencent’s 3D and video synthesis goals, as the company’s Hunyuan 3D model has already surpassed 3 million downloads on Hugging Face as of early 2026. The concentration of ex-OpenAI researchers suggests Tencent is moving toward a more cohesive multimodal stack that can natively process text, image, and 3D data in a unified framework.
Read full article at letsdatascience.com
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source