Tencent Launches Hy4 Open-Source Model

Tencent has announced Hy4, its latest large language model, positioning it among the world's leading open-source AI models. The launch matters because Tencent says the model adds visual understanding and computer use for the first time, while improving instruction following, long-context comprehension and long-horizon agentic work.

What is Tencent Hy4 and when was it announced?

According to Tencent, Hy4 was announced on November 4, 2026, as the company's latest-generation large language model. Tencent says the release represents a significant step forward in performance and capability compared with earlier versions in the Hy family.

The company also says it plans to publish the model weights and system cards in the coming days. Tencent links that planned release to its commitment to responsible open-source development, with particular attention to cybersecurity and biological safety. The announcement does not, in the material provided, give a specific date for the weights release, nor does it provide full technical specifications such as parameter count or training data size.

For readers tracking open models, the key distinction is timing: the model itself has been announced and made available through Tencent products and some third-party platforms, while the promised weights and system cards are described as forthcoming.

What new capabilities does Hy4 add?

Tencent says Hy4 introduces visual understanding for the first time in this model line. In practical terms, the company describes support for fine-grained image processing, complex chart analysis and mathematical visual reasoning, alongside broader reasoning and tool-calling abilities.

A second first is computer use. According to Tencent, Hy4 can interact with command-line interfaces and graphical user interfaces to carry out work and everyday tasks. The company says the model received specialized training on professional software including Blender and CAD, which it says improves planning and execution on long-horizon, sophisticated tasks.

Tencent also emphasizes deep model-product co-design, meaning the model was developed alongside the products it will run in. It points to office productivity, professional financial analysis, frontend and game development, and scientific research as areas where that co-design is intended to translate into real-world use.

In office settings, Tencent says Hy4 connects through WorkBuddy to workflow data from chats, emails, documents, calendars and meetings when integrated with WeCom. The company says it can track project history, identify outstanding tasks, flag meetings needing preparation and monitor approaching deadlines. A "Projects" feature for team collaboration is described as upcoming.

How did Hy4 perform in Tencent's reported evaluations?

Tencent reports several internal and partner evaluations, and readers should treat them as company-reported results rather than independent verification. On WorkBuddy, Tencent says Hy4 achieved an average performance score of 83.18 and a pass rate of 96.15 percent, ranking first in nine of twelve scenarios in the mainland Chinese market. The company says the model balances efficiency and effectiveness for visual creative tasks, 3D scene construction and web development, with improvements in document presentation, stability and usability for human-AI collaboration.

In finance, Tencent describes a blind evaluation conducted with Tencent Financial Technology involving 213 financial professionals, including 153 experts from leading external institutions, with about 70 percent having more than five years of industry experience. Across 579 real-world financial tasks, Tencent says Hy4 achieved the highest overall performance score among the open-source models compared: Hy4 at 3.04, DeepSeek-V4.1-Flash-1017 at 2.97, and Kimi K3.1 Pro and GLM-5.3 each at 2.68. Tencent also cites a 12.03 percentage-point improvement over Hy4 Preview in an internal Investment Research Agent Benchmark evaluation conducted by financial AI company Rabyte.

Tencent says targeted search optimisation, focused on retrieving up-to-date information, integrating multiple sources and verifying factual consistency, reduced the hallucination rate from 12.95 percent in Hy4 Preview to 7.2 percent, while errors from unsupported assumptions fell from 23.1 percent to 5.6 percent in internal evaluations.

For development, Tencent says Hy4 improved by 19.1 percent in GameDev and 9.8 percent in GameLogic compared with Hy4 Preview. It says the model integrates with Unreal Engine, Unity and Godot, and supports complex game logic and multiplayer implementations. For research, Tencent says Hy4 was validated with TenureTrail, which assessed 123 executable tasks across seven disciplines, with strong results reported in mathematical construction, physical modeling, environmental and earth sciences, systems engineering and socio-economic decision-making.

Tencent also says Hy4 is advancing into recursive self-improvement research. Using the Hyra 2.0 harness, it says Hy4 optimized pre-training infrastructure for smaller models in the same architecture, and that two models developed with Hy4 and Hyra 2.0 — HyCell for single-cell representations and HyPDE for physics fields — have been open-sourced as reusable research assets.

How is Hy4 being made available, and at what price?

According to Tencent, Hy4 is available across Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao, WeCom's AI Assistant (Dayuan), ima and other Tencent services. The API is also being made available through third-party platforms including OpenRouter, Hermes and OpenCode. Tencent says integration is underway for additional products and platforms, including QQ, Honor of Kings, Game of Peace, WeRead and Tencent News.

On pricing, Tencent says Hy4 keeps the same pricing as before despite performance improvements over Hy4 Preview: approximately US$0.90 per million input tokens, US$2.70 per million output tokens and US$0.045 per million cached input tokens. The company describes this as delivering greater cost performance.

A lighter variant, Hy4 Flash, is scheduled for release soon, according to Tencent, aimed at users who prioritize cost efficiency for high-value applications. For a limited period, WorkBuddy and CodeBuddy will be available for free from November 4 to 15, and Hy4 on Tencent Cloud TokenHub will be offered at 50 percent off standard pricing through December 31, 2026.

What safeguards does Tencent describe for cybersecurity and biosafety?

Tencent says safeguards are incorporated throughout training and release, with emphasis on cybersecurity and biosafety. For cybersecurity, the company says Hy4 was trained on defensive security data, prioritizing proof-of-concept development and vulnerability remediation, and incorporates reverse engineering and penetration testing for analysis. Offensive exploitation material is explicitly excluded from the training dataset, according to Tencent.

Security assessments by Tencent's Xuanwu Lab and Keen Security Lab covered cryptographic reverse engineering, binary deobfuscation, vulnerability exploitation and more. Tencent says those assessments found no significant concerns regarding Hy4's offensive cybersecurity capabilities in the areas evaluated.

On biosafety, Tencent says internal expert reviews found strong foundational biological knowledge, but more limited capability in open-ended experimental design compared with frontier models. The announcement presents these points as part of its responsible open-source approach, rather than as a full public safety evaluation. Readers who need detailed risk information will need to wait for the promised system cards.

How can readers explore this in Grafy?

The most direct Grafy connection is model comparison. Grafy Chat lets readers chat with many AI models side by side, which is a practical way to test claims about instruction following, chart analysis or long-context work against other models on the same prompt. Try Grafy Chat with a document-heavy or visual-reasoning task and compare outputs directly.

For the research angle, Grafy Reader can help readers question PDFs and papers with AI, which fits the scientific and financial use cases Tencent highlights. For frontend and game development, Grafy Code offers an AI coding studio where readers can test generated interfaces or game logic in practice. Grafy News Desk perspective: the value of an open-source release like Hy4 will depend less on headline rankings and more on whether the forthcoming weights, system cards and independent benchmarks let developers reproduce and evaluate the results.

Frequently asked questions

When does Tencent plan to release Hy4 model weights and system cards?

According to Tencent, model weights and system cards are planned for release in the coming days. The announcement frames this as part of its commitment to responsible open-source AI development, with a focus on cybersecurity and biological safety, but it does not provide a specific release date.

What price does Tencent list for Hy4 API tokens?

Tencent lists Hy4 at approximately US$0.90 per million input tokens, US$2.70 per million output tokens and US$0.045 per million cached input tokens. The company says this pricing is unchanged from before despite performance improvements over Hy4 Preview, and describes it as delivering greater cost performance.

Does Hy4 support computer use with command-line and graphical interfaces?

Yes. According to Tencent, Hy4 introduces computer-use capabilities for the first time, enabling interaction with both command-line interfaces and graphical user interfaces. Tencent says specialized training on software such as Blender and CAD improves its ability to plan and execute long-horizon tasks.

Source: Tencent — Tencent Launches Hy4, Joining the Ranks of the World's Leading Open-Source AI Models

---

Source: Hacker News — front page — Tencent Launches Hy4, One of the Leading Open-Source AI Models