Tencent Launches Hy4 with Computer-Use Capabilities and Visual Understanding

Tencent Holdings Ltd. unveiled Hy4 today, introducing computer-use functionality to its model lineup through sophisticated visual understanding, reasoning and tool-calling mechanisms designed to handle extended agentic workflows.

The release represents a substantial performance upgrade from Hy4 Preview, which debuted on August 28 with 770 billion parameters and 49 billion active parameters. Both versions support a 1 million token context window, placing them alongside frontier open-weights models from competitors like Mistral AI SAS, Beijing Moonshot Technology Co. Ltd. and Z.AI Co. Ltd.

Tencent expanded both pre-training and post-training substantially, achieving improvements across instruction following, long-context understanding, agentic long-horizon tasks, and deep research capabilities. The company identified and addressed excessive reasoning and over-verification issues that occurred during the preview version's "thinking" processes.

Workplace and Productivity Focus

The model emphasizes practical applications for workplace collaboration, user experience and professional contexts. Tencent optimized long-context learning to process complex information spanning web development, document summarization, visual creative work, and presentation generation.

Front-end and game development capabilities received particular attention, addressing high-complexity scenarios involving intricate business logic, state management and orchestration. Many AI models struggle to produce stateful, visually polished and fully functional user interfaces, an area where Hy4 aims to excel.

Game Development and Integration

Professional game developers, industry engineers and independent developers frequently incorporate AI models into their workflows. A recent trend of developers reproducing, reverse engineering and cloning older and semi-recent video games prompted Anthropic PBC to restrict Claude Code and its underlying Opus 5.5 model from assisting with such activities.

Tencent integrated Hy4 deeply with major engines including Unreal Engine, Unity and Godot, enabling support for complex game logic and multiplayer systems. Compared to the preview version, Hy4 achieved a 19.1 percent improvement on the GameDev bench and a 9.8 percent improvement on GameLogic benchmarks, though Tencent did not disclose the underlying baseline figures.

Scientific Reasoning and Self-Improvement

Tencent advanced its Recursive Self-Improvement methodology to strengthen research and development capabilities. The Hyra 2.0 harness, an autonomous AI research agent system, enabled the company to optimize pre-training infrastructure for smaller models sharing the same architecture.

Availability and Pricing

Hy4 is accessible through Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao, WeCom's AI Assistant, Ima and additional platforms. Third-party access is available via OpenRouter, Hermes and OpenCode.

Despite significant enhancements over Hy4 Preview, Tencent maintained pricing at approximately $0.90 per million input tokens, $2.70 per million output tokens, and $0.045 per million cached input tokens.

A Flash version is coming soon to provide greater cost efficiency. Tencent also announced plans to release model weights and system cards shortly as part of its approach to responsible open-source AI development.