Tencent Hy4 AI model outperforms GPT in computer control benchmark

Tencent has officially introduced Hy4, marking the first model in the Tencent Hy product line to incorporate visual input alongside the ability to control computers through both command-line terminals and graphical user interfaces.

Key highlights of the Tencent Hy4 model: - Integrated visual input and computer control via terminal and GUI. - Massive context window reaching up to 1 million tokens. - Optimized for software development, workflow automation, and UI creation. - Outperformed GPT models in the OSWorld 2.1 benchmark.

Capabilities and Technical Performance

The Hy4 model is specifically designed for software engineering, automating complex workflows, document processing, and user interface design. With a context window expanding up to 1 million tokens, the model can handle extensive volumes of data and code repositories in a single request.

During evaluations on OSWorld 2.1—a benchmark measuring an AI's ability to execute tasks within a real operating system environment—Hy4 successfully outperformed GPT models and came close to matching the performance of industry leaders like Claude, although gaps remain in other testing categories.

Implications for the Tech Industry

The emergence of agent models capable of interacting directly with operating system interfaces and processing massive contexts highlights the rapid evolution of multimodal AI. For developers and enterprises, such tools pave the way for deeper workflow automation and smarter computing assistants.

As Tencent continues to scale its AI ecosystem, innovations like Hy4 demonstrate the growing capability of Asian tech giants to challenge established Western models in complex reasoning and autonomous task execution.