Tencent Releases AI Image Model to Catch ByteDance, Alibaba

Tencent Holdings Ltd. launched its Hy Image 3.5 Preview image-generation model on Tuesday, the same day rival Alibaba Group Holding Ltd. opened its AI conference, and said the model matches ByteDance Ltd.'s Seedream 5.0 Pro while edging past Google's Nano Banana Pro and Alibaba's Qwen-Image-3.0 Pro. Tencent is integrating Hy Image 3.5 Preview across its products, including the Yuanbao chatbot and film-editing and design tools, and its Hong Kong shares gained more than 7%. The release follows Tencent's hiring of former OpenAI researchers Yao Shunyu, now chief AI scientist, and Tian Yonglong, who joined in July to lead vision-language models under the Hunyuan team.

Bloomberg -- Tencent Holdings Ltd. launched its latest and best image-generation model on the same day as rival Alibaba Group Holding Ltd. kicked off its AI conference, signaling intent to catch up to industry leaders like ByteDance Ltd. Most Read from Bloomberg The Hy Image 3.5 Preview model offers a performance boost over its predecessor and is being integrated across Tencent products, including the Yuanbao chatbot and film-editing and design tools, the company said in a statement Tuesday. The release shows Tencent accelerating its exploration of applications beyond text, while working toward its vision of building AI that performs tasks for the billion-plus users of its WeChat ecosystem. Meta Platforms Inc.'s agent-centric Muse this week shot to the top of download charts on mobile app stores — underscoring interest in an agentic AI that can tailor content for users and help them carry out tasks like shopping and hailing rides. Tencent's shares gained more than 7% in Hong Kong. "Some investors are drawing parallels between Tencent and Meta, particularly given WeChat's unique social ecosystem and its potential to support a personal AI assistant at scale," said Gary Tan, portfolio manager at Allspring Global Investments. The Chinese social media and gaming conglomerate has lagged Big Tech peers in delivering boundary-pushing large language models. It recently hired former OpenAI researcher Yao Shunyu, who overhauled the company's AI infrastructure and benchmarking standards. Yao, who now serves as chief AI scientist, has publicly criticized training models solely for the sake of leaderboard scores, saying Tencent will instead focus on product integration and solving real-world problems. The company has also continued to attract top talent since hiring Yao. Tian Yonglong, a former OpenAI researcher specializing in computer vision, joined Tencent in July and now leads vision-language models under the Hunyuan team. Shenzhen-based Tencent tested the Hy Image 3.5 Preview among hundreds of in-house designers, and said that the model achieved results on par with ByteDance's Seedream 5.0 Pro. The new Hy Image model is slightly better than Nano Banana Pro from Alphabet Inc.'s Google and Qwen-Image-3.0 Pro from Alibaba, the company added. Tencent employs thousands of game designers and artists, who may benefit from the tool, though it didn't quantify its quality claims.