Hy4 introduces visual understanding capabilities for the first time; introduces computer-use capabilities for the first time; provides major improvements in instruction following, long-context comprehension and long-horizon agentic tasks; and delivers stronger performance across diverse application, according to the release.
The new capabilities unlock new opportunities for real-world productivity, the release said. Tencent highlighted office productivity and collaboration, frontend and game development and scientific research.
“Hy4 has also been optimized for professional financial analysis,” Tencent said in the release. “By training on thousands of specialized workspaces and tens of thousands of high-quality tasks derived from real-world scenarios, the model gains a deep understanding of expert workflows in fund research, securities analysis and quantitative investment strategies.”
Tencent also highlighted cybersecurity and biosafety safeguards the company built into Hy4. The model is trained on defensive security data but not on offensive materials related to exploitation, the release said. In terms of biosafety, Hy4’s capabilities in experimental design are limited in comparison to frontier models, per the release.
Tencent will soon release a Hy4 Flash model that is designed for greater cost-efficiency and high-value applications, according to the release.
It was reported Wednesday that Oracle terminated a $7 billion agreement to provide advanced AI computing capacity to Tencent, after U.S. government officials raised concerns about this arrangement with the Chinese company.
The proposed five-year contract would have allowed Tencent to use sophisticated AI processors housed in Oracle-operated data centers across Southeast Asia and would have provided Tencent with access to about 100,000 advanced AI chips that are difficult to obtain within mainland China because of restrictions imposed by Washington and Beijing.
In October, Tencent’s AI Unit, Tencent Hunyuan (Tencent Hy), said that rogue AI agent activity covered in media reports at the time originated from evaluations of third-party models, not the company’s own models.
It was reported at the time that this appeared to be the first known incident in which a major Chinese company acknowledged actions by rogue agents.
