Tencent Hunyuan Releases Hy4, Entering Top Open-Source Model Ranks

Tencent Hunyuan has just released Hy4, its latest-generation large language model. With another comprehensive leap in model capabilities, Hy has for the first time joined the ranks of the strongest open-source models. Hy4 will soon open-source its model weights, publish a model system card, and actively advance more responsible open-sourcing, with a focus on risks such as cyber and biosecurity.

Compared with Hy4 preview, Hy4 significantly expands the data and training scale in both pre-training and post-training, introduces visual understanding capabilities for the first time, and substantially improves comprehensive performance in instruction following, long-document understanding, and long-horizon agents.

Through visual understanding, reasoning, and tool-calling capabilities, Hy4 can complete tasks such as fine-grained image processing, complex chart analysis, and mathematical visual reasoning.

Hy4 also supports Computer Use for the first time, enabling it to handle various tasks in work and life by combining command-line and graphical interfaces. For professional software such as Blender and CAD as well as cross-application long-horizon tasks, Hunyuan conducted specialized training that greatly improves the planning and execution capabilities for professional tasks.

Determining geographic location and verifying information from casually taken photos

Deep Co-Design of Models and Products, Continuously Unlocking Real Productivity Scenarios

Thanks to the comprehensive improvement in the model's foundational capabilities and the deep Co-Design between the model and products, Hy4 has unlocked more real productivity scenarios.

In office scenarios, Hy4 comprehensively improves the user experience of WorkBuddy. Evaluation results show that Hy4 achieves an average performance score of 83.18, a task pass rate of 96.15%, and ranks first in China in 9 out of 12 scenarios. In practical experience, Hy4 balances effectiveness and efficiency in tasks such as visual creativity, 3D scene construction, and web development. At the same time, for WorkBuddy's distinctive human-machine dual-writing feature, Hy4 shows clear improvements in document aesthetics, stability, and usability.

With the continuous improvement of Hy4's in-context learning capabilities, the model can better understand multi-source complex information and become users' long-term personal assistant. With WorkBuddy's WeCom suite, Hy4 can connect daily work information such as chat records, emails, documents, schedules, and meetings. When multiple projects are advancing in parallel, Hy4 can clarify the advancement history and latest progress of each project, identify tasks to follow up on, meetings to prepare for, and items approaching deadlines, promptly remind users, and help prioritize. In addition, the team collaboration feature powered by Hy4 will be launched in WorkBuddy with the new version of "Projects."

Invoice organization and financial invoice auditing after a business trip

Hy4 has focused on strengthening professional financial analysis capabilities. Based on real financial scenarios, it built thousands of workspaces and tens of thousands of high-quality tasks, allowing the model to fully learn the professional processes and standards in areas such as fund research, securities analysis, quantitative strategies, asset allocation, corporate finance, and fiscal and tax accounting. Hunyuan, together with Tencent FinTech (FiT), conducted a financial public test with 213 experts, 153 of whom came from leading external financial institutions, and 70% of whom have more than five years of work experience. Results based on 579 real financial tasks show that Hy4 ranks first among open-source models in overall performance (public test scores: Hy4 3.04, Kimi K3.1 Pro 2.68, GLM-5.3 2.68, DeepSeek-V4.1-Flash-1017 2.97). In addition, in the internal evaluation IRAB of the financial agent company Xuntu Technology, Hy4 improved by 12.03 percentage points over Hy4 preview.

To improve practicality in finance and various other office scenarios, Hunyuan also launched specialized search optimization for time-sensitive retrieval, multi-source information integration, and factual consistency verification. In self-built tests, Hy4's hallucination rate dropped from 12.95% in the preview version to 7.2%, and the error rate caused by guessing dropped from 23.1% to 5.6%.

In front-end and game development scenarios, Hunyuan carried out specialized optimization for complex front-end interactions, business logic, state management, and front-end/back-end integration, so that the interfaces generated by the model are not only attractive but also fully functional and usable. Hunyuan collaborated with professional game developers at Tencent IEG to continuously optimize data and evaluation quality, focusing on improving capabilities such as game feature implementation, gameplay logic, state understanding, and asset processing, achieving deep adaptation to mainstream engines such as UE, Unity, and Godot, and beginning to expand to multiplayer online game development. Compared with Hy4 preview, Hy4 improved by 19.1% and 9.8% in the GameDev and GameLogic evaluations, respectively.

Retro QQ game arcade

In the field of scientific research, Hy4 combines complex reasoning, scientific programming, and long-horizon task execution to further enhance its ability to advance research end to end. In the internal TenureTrail evaluation, which covers 123 executable research tasks across 7 major disciplines, Hy4 demonstrates solid research capabilities in areas such as mathematical construction, physical modeling, Earth and environment, systems engineering, and socioeconomic decision-making.

Hy4 is actively preparing for the next stage of recursive self-improvement (RSI), enhancing capabilities across various model R&D tasks. Combined with Hyra2.0, Hy4 optimized the pre-training infrastructure for internal small models of the same architecture and automatically trained the Jev model, which outperforms the official model in multiple evaluations. At the same time, Hunyuan has also open-sourced the single-cell representation foundation model HyCell and the physical field foundation model HyPDE, developed by Hy4 in collaboration with Hyra2.0, hoping to turn the results of model-assisted scientific research into reusable tools for the community.

Focusing on Cyber and Biosecurity, Continuously Advancing Responsible Open-Sourcing

In response to the potential security risks of models in areas such as cyber and biology, Hunyuan conducts training, open-sourcing, and release in a more responsible manner.

In terms of cybersecurity, Hy4 includes only defensive cybersecurity data related to PoC generation and vulnerability remediation in training, while removing offensive training content such as vulnerability exploitation, reverse engineering, and penetration, striving to avoid improvements in potentially offensive capabilities. Hunyuan organized security experts from Tencent Xuanwu and Keen Labs to conduct audits in areas such as cryptanalysis, reverse engineering, and vulnerability exploitation. The results show that Hy4 does not exhibit prominent risks in offensive capabilities such as reverse solving of encryption algorithms, adversarial binary obfuscation, and vulnerability exploitation.

In terms of biosecurity, there is a clear gap between Hy4's open-ended experimental design capability and that of frontier models (Hy-BioExperimentBench: 30.4/100). Hunyuan organized internal experts to audit the trajectories and found that the model has a strong grasp of basic biological knowledge, but its ability to translate that knowledge into executable experimental protocols remains limited, and it has not been found to significantly lower the barrier to carrying out biochemically harmful behaviors.

Hy4 completes 3D reconstruction of a 7-million-cell mouse embryo in 1.5 hours on a single machine

Continuing to pursue ultimate cost-effectiveness, Hy4 flash to launch soon

Currently, Hy4 has been launched on products including Tencent Cloud Tokenhub, WorkBuddy/CodeBuddy, Yuanbao, WeCom Intelligent Assistant Dayuan, and ima, with its API also integrated into platforms such as OpenRouter, Hermes, and OpenCode. Tencent products such as QQ, Honor of Kings, Game for Peace, WeRead, and Tencent News are also continuing to integrate it.

Tencent Hunyuan takes "model practicality" as its core principle, continuously pursuing the upper limits of intelligence while providing users with cost-effective services. While Hy4's model performance substantially surpasses that of Hy4 preview, its pricing remains unchanged: 6 yuan per million input tokens, 18 yuan per million output tokens, and 0.3 yuan per million tokens for cache hits.

Hy4 flash will also be released soon. Building on Hy4's capabilities, Hy4 flash significantly reduces task costs, providing users with an extremely cost-effective model.

To collect more user feedback and continuously improve model capabilities, WorkBuddy/CodeBuddy will offer users a certain amount of free usage from November 4 to November 15. In addition, from now until December 31, 2026, Hy4 will be billed at 50% of the list price on Tencent Cloud TokenHub.

Since rebuilding its infrastructure in February this year, the Tencent Hunyuan large model has released a major version on average every two months, first opening it in Preview form and then releasing the official version after optimizing the model based on user feedback.

At the same time, Hunyuan continues to scale up in both parameter size and data size. The Hy3 series has fewer than 300B parameters, while the Hy4 series reaches 770B. It is reported that the next-generation Hy5 will be further expanded in scale.

The Hunyuan team stated that although Hy4 is already one of the strongest open-source models currently available, there is still a considerable gap between it and top closed-source models. Hunyuan will continue to steadily expand the scale of pre-training and post-training, striving to narrow the gap with top closed-source models.

Open in WeChat