Alibaba teases 10-trillion-parameter model, debuts ‘China’s most powerful’ AI chip

Tencent unveiled its latest image-generation model, Hy Image 3.5 Preview, on the same day as ByteDance and Alibaba Group launched their AI conference, indicating a desire to keep pace with industry giants. Hy Image 3.5 Preview outperforms its predecessor and is being incorporated into various Tencent products, such as the Yuanbao chatbot and design tools.

The release demonstrates Tencent's increased focus on expanding AI applications beyond text, with the goal of fulfilling the needs of WeChat's billion-plus users. Meta Platforms' agent-centric Muse recently topped mobile app store download charts, highlighting the growing interest in AI agents that can personalize content and complete tasks for users.

Tencent's shares rose by more than 7% in Hong Kong, with analysts drawing comparisons to Meta, particularly due to WeChat's social ecosystem and potential for a personal AI assistant on a wide scale. The company recently appointed former OpenAI researcher Yao Shunyu as chief AI scientist, who has emphasized product integration and solving real-world problems over merely achieving top leaderboard scores.

Additionally, Tencent has attracted top talent, including Tian Yonglong, a former OpenAI researcher specializing in computer vision, who now leads vision-language models under the Hunyuan team. In-house testing of Hy Image 3.5 Preview showed results comparable to ByteDance's Seedream 5.0 Pro, and slightly surpassed Nano Banana Pro from Google and Qwen-Image-3.0 Pro from Alibaba.

With thousands of game designers and artists at Tencent, the new image model could significantly benefit their workforce, although the company did not specify the extent of these potential enhancements.