Tencent Launches Hunyuan Hy3 AI Model Preview and Considers DeepSeek Investment
Tencent's Hy3 AI model preview boosts reasoning and coding, while DeepSeek drops V4 Flash and Pro.
Tencent has launched the preview of its Hunyuan Hy3 AI model, its first significant update since recruiting Yao Shunyu, a leading AI scientist from OpenAI. The Hy3 model offers improvements in complex reasoning and coding, positioning Tencent to compete with other Chinese AI developers like ByteDance, Alibaba, and DeepSeek. Tencent is also restructuring its research team to enhance training data quality and aims to double its AI investment to over $5 billion this year, backing startups like Moonshot AI and StepFun to boost its cloud computing division.
Not to be outdone, DeepSeek has announced its V4 Flash and V4 Pro series, the latest versions of its LLM model. DeepSeek, which became a sensation in January 2025 with its R1 model, now offers advances in reasoning and agentic tasks. A new Hybrid Attention Architecture improves the AI's ability to remember queries across long conversations. Both companies are part of China's broader push for open-source AI alternatives to US players like OpenAI and Meta.
- Tencent's Hy3 model preview enhances complex reasoning and coding, its first major update since hiring OpenAI scientist Yao Shunyu.
- Tencent plans to double AI investment to over $5 billion in 2025, backing startups like Moonshot AI and StepFun.
- DeepSeek's V4 Flash and V4 Pro series introduce Hybrid Attention Architecture for better long-context memory and improved reasoning and agentic tasks.
Why It Matters
China's AI race heats up with Tencent and DeepSeek releasing competitive models, challenging US dominance in open-source AI.