Tencent released Hy4 on November 4, introducing vision and computer-use capabilities to its latest flagship large language model. The release delivers broad improvements over Hy4 preview and brings Hy4 into the ranks of the world’s most capable open-weight models for the first time.
Tencent will release Hy4’s model weights shortly and publish a system card for the model.
From visual understanding to computer use
Hy4 significantly expands data and training scale across both pre-training and post-training. Alongside its first vision capabilities, the model delivers comprehensive improvements in instruction following, long-context understanding and long-horizon agentic capabilities.
Its vision capabilities bring together visual understanding, reasoning and tool use, allowing Hy4 to interpret dense charts, parse documents and analyze information across multiple images. It can also combine visual information for multi-step analysis and deduction, extending beyond basic visual understanding into visual reasoning.
Building on these capabilities, Hy4 supports computer use for the first time, operating across both the command line and graphical interfaces to handle a wide range of work and everyday tasks. Tencent’s Hy team conducted targeted training on specialized software, including Blender and CAD applications, as well as long-horizon tasks spanning multiple applications. This further improves the model’s ability to plan and execute professional workflows.
Co-design connects model capabilities with real workflows
In office scenarios, Hy4 improves the WorkBuddy user experience across the board. A WorkBuddy-focused evaluation recorded an average quality score of 83.18 and a pass rate of 96.15%. In real-world use, Hy4 balances quality and efficiency on tasks including visual design, 3D scene construction and web development.
Professional financial analysis also receives a major upgrade, supported by thousands of workspaces and more than 10,000 high-quality tasks built from real-world finance scenarios. These cover fund research, securities analysis and quantitative strategy work.
In a finance-focused blind evaluation, 153 of the 213 experts came from leading financial institutions outside Tencent, and roughly seven in ten had more than five years of experience. Across 579 real-world finance tasks, Hy4 ranked first among the open-weight models evaluated on overall performance.
For front-end development, the Hy team has made targeted optimizations in complex interactions, business logic, state management and front-end/back-end integration. These improvements help the model generate interfaces that are visually appealing, functionally complete and usable.
In game development, Hy has worked with professional developers at Tencent’s IEG to continuously improve data and evaluation quality. The work focuses on game feature implementation, gameplay logic, state understanding and asset processing. Hy4 has achieved deep adaptation to mainstream engines including UE, Unity and Godot, and has begun extending into online multiplayer game development. Compared with Hy4 preview, it improved by 19.1% on GameDev and 9.8% on GameLogic.
In scientific research, Hy4 combines complex reasoning, scientific programming and long-horizon task execution to strengthen its ability to advance research workflows end to end. In TenureTrail, an internal evaluation spanning seven disciplines and 123 executable research tasks, Hy4 demonstrated solid research capabilities in areas including mathematical construction, physical modeling, earth and environmental sciences, systems engineering and socio-economic decision-making.
Tencent believes model development should extend beyond the laboratory. The Hy team works with product teams, particularly WorkBuddy, on joint development, or Co-Design, within real-world enterprise workflows.
Since launching in March 2026, WorkBuddy has released more than 100 major version updates. Its token consumption has grown 50-fold compared with the initial launch period in March, and it has completed more than 9 billion task requests to date.
WorkBuddy is now China’s largest AI agent for office productivity by user base. Its enterprise edition has been deployed across more than 50 industries, serving 50% of China’s top 500 enterprises and 60% of its top 100 enterprises.
Responsible development and open-weight release
To address potential risks in areas including cybersecurity and biology, Hy has taken a more responsible approach across training, open-weight release and launch.
For cybersecurity, Hy4 was trained only on defensive data related to proof-of-concept (PoC) generation and vulnerability patching. Offensive material covering exploitation, reverse engineering and penetration testing was excluded to avoid any potential uplift in offensive capabilities.
The Hy team brought together in-house security experts from Tencent Xuanwu Lab and Keen Lab to audit areas including cryptanalysis, reverse engineering and exploitation. The results showed no notable risk in Hy4’s offensive capabilities, including reverse-solving cryptographic algorithms, countering binary obfuscation and exploiting vulnerabilities.
For biological safety, Hy4’s open-ended experiment design remains well behind frontier models, scoring 30.4/100 on Hy-BioExperimentBench. Internal experts audited the model’s trajectories and found a solid grasp of foundational biology but limited ability to translate that knowledge into executable experimental plans. They found no evidence that Hy4 can meaningfully lower the barrier to carrying out biochemical harm.
Availability and Pricing
Hy4 is now live across Tencent’s product portfolio. Its API is also available through third-party platforms including OpenRouter, Hermes and OpenCode.
Despite the substantial capability improvements, Hy4 retains Hy4 preview’s pricing: $0.834 per million input tokens, $2.501 per million output tokens and $0.042 per million tokens on cache hits. Tencent Hy remains committed to delivering cost-efficient AI services.
Hy4 flash will also be released shortly, retaining Hy4’s capabilities while substantially reducing the cost per task, with a focus on maximum price-performance.
To gather more user feedback and continuously improve model capabilities, WorkBuddy and CodeBuddy will offer users a limited quota of free usage from November 4 to November 15. From November 4 through December 31, 2026, Hy4 will also be billed at 50% of its list price on Tencent Cloud TokenHub.
In model R&D, Tencent Hy is preparing for a future phase of research into recursive self-improvement (RSI), focusing on using models to accelerate model development and explore frontier problems.
Since rebuilding its AI infrastructure in February 2026, Tencent Hy has shipped a major model roughly every two months. Its models have scaled from under 300 billion parameters in Hy3 to 770 billion in Hy4, with Hy5 expected to go further.
The Tencent Hy team said that while Hy4 is already among the most capable open-weight models available today, a considerable gap remains between it and leading closed-source models. Hy will continue scaling pre-training and post-training in a steady, measured way to close that gap.
