Tencent's Hy4 adds vision and computer use, ranks first among open-weight models in blind finance evaluation

Tencent's Hy4 adds vision and computer use and joins the top tier of open-weight models. On a WorkBuddy-focused evaluation, Hy4 recorded an average quality score of 83.18 and a pass rate of 96.15%. WorkBuddy is Tencent's AI agent for office work.

Pricing, context window and parameter count for Hy4 were not disclosed in the article.

Built with product teams

Tencent says the Hy team works with product teams on joint development (Co-Design) inside real enterprise workflows, with a particular focus on WorkBuddy. In real-world use, Hy4 balances quality and efficiency on tasks such as visual design, 3D scene construction and web development.

WorkBuddy launched in March 2026 and has released more than 100 major version updates since. Token consumption has grown 50-fold compared with the initial launch period in March, and the agent has completed more than 9 billion task requests. It is now the largest AI agent for office productivity in China by user base. Its enterprise edition is deployed across more than 50 industries and serves 50% of China's top 500 enterprises and 60% of its top 100 enterprises.

Finance analysis

Tencent built thousands of workspaces and more than 10,000 high-quality tasks from real finance scenarios, covering fund research, securities analysis and quantitative strategy work. In a blind evaluation by 213 finance experts, Hy4 ranked first in overall performance among the open-weight models evaluated across 579 real-world finance tasks. Of the 213 experts, 153 came from leading financial institutions outside Tencent, and roughly seven in ten had more than five years of experience.

Front-end and game development

The Hy team made targeted optimizations in complex front-end interactions, business logic, state management and front-end/back-end integration. The goal is interfaces that are functionally complete as well as visually appealing.

Tencent worked with professional game developers at Tencent Interactive Entertainment Group (IEG) to improve data and evaluation quality. The focus areas were game feature implementation, gameplay logic, state understanding and asset processing. Hy4 is adapted to Unreal Engine, Unity and Godot, and has begun extending into online multiplayer game development. Compared with Hy4 preview, Hy4 improved by 19.1% on the GameDev evaluation and 9.8% on the GameLogic evaluation.

Scientific research

Hy4 combines complex reasoning, scientific programming and long-horizon task execution. On TenureTrail, an internal evaluation spanning seven disciplines and 123 executable research tasks, Hy4 showed solid research capabilities in mathematical construction, physical modeling, earth and environmental sciences, systems engineering and socio-economic decision-making.

What this means

The evidence Tencent offers for Hy4 centers on applied workloads rather than general leaderboards: office tasks in WorkBuddy, finance analysis, game development and research execution. The finance result carries the most weight because it comes from a blind evaluation with a majority of evaluators from outside Tencent, though it ranks Hy4 only against the open-weight models evaluated. The WorkBuddy scores and the TenureTrail evaluation come from Tencent's own test setups. Builders considering Hy4 for agentic office or game-engine work should treat the WorkBuddy figures as product-specific results, and check the model's vision and computer-use behavior against their own workflows.