# Tencent Releases Hy4, a High-Performance Open-Source AI Model

Tencent's new Hy4 model introduces advanced visual reasoning and computer-use capabilities for professional workflows.

- **Source:** [Hacker News: Front Page](https://brifing.si/source/b07aa772-b1a0-4a76-aa18-719c6f9e2dde)
- **Published:** 2026-11-05
- **Topic:** [technology](https://brifing.si/c/technology)
- **Original article:** <https://www.tencent.com/tencent-launches-hy4-joining-the-ranks-of-the-worlds-leading-open-source-ai-models/>
- **On Brifing:** https://brifing.si/s/9a10e303-1c4f-4948-84c4-1b45ffdc915e

## Launch of Hy4 Model

Tencent officially released Hy4, its latest large language model, positioning it as a top-tier open-source competitor. The company plans to release model weights and system cards shortly. This release emphasizes responsible development, specifically focusing on cybersecurity and biological safety protocols to ensure the model remains secure for diverse professional applications.

## Visual and Agentic Capabilities

Hy4 introduces significant visual understanding, allowing for complex chart analysis and fine-grained image processing. The model also features new computer-use capabilities, enabling interaction with command-line and graphical interfaces. Trained on professional software like CAD and Blender, it excels at planning and executing long-horizon tasks, marking a major leap in agentic performance.

## Office Productivity Integration

Integrated into Tencent WorkBuddy, the model achieved an 83.18 performance score in internal benchmarks. It ranks first in nine out of twelve scenarios within the Chinese market. By connecting with WeCom, Hy4 tracks project history, manages emails, and flags upcoming deadlines, effectively streamlining collaborative workflows for teams using the platform.

## Financial Analysis Performance

In a blind evaluation involving 213 financial professionals, Hy4 outperformed competitors like DeepSeek-V4.1 and Kimi K3.1 Pro. The assessment covered 579 real-world tasks, including fund research and quantitative strategies. Hy4 achieved a score of 3.04, demonstrating superior accuracy in complex financial reasoning compared to other leading open-source models currently available.

## Game Development Advancements

Tencent optimized Hy4 for high-complexity tasks, including business logic and state management in game development. Through integration with engines like Unreal Engine, Unity, and Godot, the model improved by 19.1 percent in GameDev benchmarks. This ensures that generated interfaces are fully functional and visually polished for professional developers working on multiplayer projects.

## Scientific Research Support

Hy4 supports end-to-end research workflows by combining scientific programming with advanced reasoning. Validated across seven disciplines, including physical modeling and socio-economic decision-making, it provides robust assets for discovery. Tencent also open-sourced two foundational models, HyCell and HyPDE, specifically designed to assist the research community with single-cell representations and physics-based field simulations.

## Safety and Hallucination Reduction

Tencent implemented targeted search optimization to improve factual consistency and reduce errors. Internal evaluations show the hallucination rate dropped from 12.95 percent to 7.2 percent, while unsupported assumptions fell to 5.6 percent. Security assessments by Xuanwu and Keen labs confirmed no significant concerns regarding offensive cybersecurity capabilities within the model's current training parameters.

## Market Availability and Pricing

Hy4 is available across Tencent Cloud TokenHub and various third-party platforms like OpenRouter. Despite performance gains, pricing remains stable at 0.90 dollars per million input tokens. Tencent is offering a 50 percent discount on TokenHub through December 2026, while WorkBuddy and CodeBuddy will be available for free until November 15.

## Sources covering this story

- https://www.tencent.com/tencent-launches-hy4-joining-the-ranks-of-the-worlds-leading-open-source-ai-models/ — 

---

_Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._
