What is Hy4 preview
Hy4 preview is an open-source flagship large model launched by Tencent Hunyuan, with 770B total parameters, 49B activated parameters, and a context length of up to 1M. Through deep co-construction with experts in Tencent's internal software engineering, gaming, finance, security, and other fields, the model performs excellently on real productivity tasks such as code development, office analysis, game production, and scientific research, and firmly ranks in the first tier of open-source models in multiple authoritative benchmark tests and 203 blind engineering task tests. In the scientific research field, it has achieved substantive breakthroughs in molecular dynamics simulations, quantum device design, and a century-old geometry problem.
Main features of Hy4 preview
- Software engineering: Enhances the understanding, planning, debugging, and verification capabilities for long-horizon development tasks, further improving the visual aesthetics and interaction quality of front-end development.
- Office analysis: Significantly improves the understanding of complex office environments and financial analysis capabilities, completing the full process from information processing to the delivery of documents, spreadsheets, and presentations.
- Game development: Enhances the ability to directly generate playable prototypes from a single requirement, is proficient in using game engines, and supports continuous improvement of complex game projects through multi-turn interaction.
- Scientific research: Significantly improves the understanding, reasoning, and solving capabilities for complex scientific research problems, with substantial progress in scenarios such as AI R&D, molecular dynamics simulations, condensed matter physics, and fundamental mathematics.
- Inference optimization: Autonomously analyzes inference system bottlenecks and carries out multiple rounds of optimization around directions such as operator fusion and communication optimization, increasing end-to-end throughput by 31.8% compared with the baseline.
- Experiment coordination: Manages multiple Codex Sessions to conduct experiments like a researcher, outperforming independent Codex exploration on all 8 evaluations, demonstrating the ability to judge direction, organize experiments, and continuously iterate in complex R&D tasks.
Technical principles of Hy4 preview
- Software engineering: Hy4 preview significantly enhances the understanding, planning, debugging, and verification capabilities for long-horizon development tasks. In front-end development of large projects, it can proactively improve visual aesthetics and interaction quality, and the generated code can run directly and be continuously iterated in real engineering environments.
- Office analysis: Through high-quality data co-construction with financial experts, the model significantly improves its understanding of complex office environments, can handle tasks such as cross-file collaboration and data analysis, and completes the full office workflow from raw information processing to the final delivery of documents, spreadsheets, and presentations, such as automatically reviewing 24 reimbursement forms and cross-comparing policy documents.
- Game development: Hy4 preview has the ability to "generate a playable prototype from one sentence" and is proficient in using mainstream game engines such as Unreal 5 and Unity. Developers can continuously improve complex game projects through multi-turn natural language interaction, covering the complete development chain including scene construction, gameplay logic, character animation blend trees, and sound effect systems.
- Scientific research: In scientific research scenarios, the model demonstrates deep understanding, reasoning, and solving capabilities for complex problems. With the Hyra agent, it speeds up machine learning force field molecular dynamics simulations by 2.0 times, reduces the leakage rate from 48.2% to 4.8% in low-temperature quantum transport device design, and advances the volume lower bound to 0.41104 on the century-old classical geometry problem 3D Blaschke-Lebesgue.
- Inference optimization: The Hy4 preview autonomously analyzes reasoning system performance bottlenecks and conducts multiple rounds of iterative optimization around directions such as operator fusion and communication optimization. End-to-end service throughput improves by 31.8% compared with the baseline, achieving stable gains under different context lengths and concurrency levels, validating the model's ability to autonomously locate bottlenecks and continuously optimize infrastructure.
Follow WeChat and reply "open source" to join theAI Open Source Project Communication Group
How to use Hy4 preview
- Open Source Download: Obtain model weights and code through HuggingFace, GitHub, ModelScope, or GitCode, supporting local deployment and secondary development.
- API Access: Connect to Tencent Cloud TokenHub or OpenRouter, and directly call the model interface at the pricing of 6 yuan per million input Tokens and 18 yuan per million output Tokens.
- Product Experience: Directly experience Hy4 preview's office and programming capabilities in Tencent products such as WorkBuddy、CodeBuddy、Yuanbao、ima .
- Online Demo: Visit the model's online experience URL https://aistudio.tencent.com/，无需配置环境即可在线与模型交互。
Core Advantages of Hy4 preview
- Leading Scale: With 770B total parameters, 49B activated parameters, and a 1M context length, it firmly ranks in the first tier of open-source models in terms of model size and long-text processing capability.
- Real Productivity Orientation: Deeply co-built with top experts in Tencent's internal software engineering, gaming, finance, security, and other fields, focusing on real scenarios such as coding, office work, and gaming rather than merely chasing benchmarks.
- Scientific Research Breakthrough Capability: Working with Hyra, it accelerates molecular dynamics simulations by 2.0x, reduces quantum device leakage rates to 4.8%, and advances the volume lower bound of the century-old geometric problem 3D Blaschke-Lebesgue to 0.41104.
- Self-Optimizing Reasoning: It can autonomously analyze system bottlenecks and continuously iterate around operator fusion and communication optimization, improving end-to-end service throughput by 31.8% compared with the baseline.
- Intelligent Experiment Coordination: It manages multiple Codex Sessions in parallel to conduct experiments like a researcher, and in small-model post-training tasks, all 8 evaluations outperform independent exploration by Codex.
- Leading in Expert Blind Tests: In blind tests of 203 engineering tasks by 163 internal experts, Hy4 preview scored an average of 2.99/4.00, slightly outperforming GLM-5.3 and Kimi K3.
Hy4 preview Project Address
- Project Official Website：https://hy.tencent.com/research/hy4-preview
- GitHub Repository：https://github.com/Tencent-Hunyuan/Hy4-preview
- HuggingFace Model Library：https://huggingface.co/tencent/Hy4-preview
Comparison of Hy4 preview with Similar Competitors
| Comparison Dimension | Hy4 preview | DeepSeek-V4 Pro |
|---|---|---|
| Total Parameters / Activated Parameters | 770B / 49B | Approximately 671B / 37B |
| Context Length | 1M | 128K |
| Model Architecture | MoE (Mixture of Experts) | MoE (Mixture of Experts) |
| Open Source Status | Fully open source (weights + code) | Open-source weights, open-source inference code |
| Core Positioning | Real productivity (coding, office work, gaming, scientific research) | General reasoning and coding |
| Scientific Research Capability | Working with Hyra, it achieves breakthroughs in molecular dynamics, quantum devices, and mathematical proofs | Strong general reasoning; scientific research toolchains need to be built independently |
| Product Collaboration | Deeply integrated with WorkBuddy, CodeBuddy, Yuanbao, ima | Mainly relies on third-party integrations and self-deployment |
| API pricing (input/output) | 6 yuan / 18 yuan (per million Tokens) | Approximately 2 yuan / 8 yuan (per million Tokens) |
| Internal blind test performance | Average score 2.99/4.00, slightly better than GLM-5.3 and Kimi K3 | Excellent benchmark performance, but no public comparable engineering blind tests |
Application scenarios of Hy4 preview
- Intelligent code development: Supports long-horizon project understanding, front-end visual aesthetic optimization, and debugging verification, and can work with CodeBuddy to complete the full-process development from requirements to runnable code.
- Complex office automation: Handles cross-file data analysis, financial auditing, and policy comparison, automatically generates documents, spreadsheets, and presentations, and has already been deployed in office products such as WorkBuddy.
- Rapid game prototype construction: A single natural language requirement can generate a playable prototype, with proficiency in Unity / Unreal 5 engines, and supports multiple rounds of interaction to continuously refine levels, characters, and gameplay.
- Scientific research acceleration: Works with the Hyra agent to carry out molecular dynamics simulations, quantum device design, and mathematical theorem proving, autonomously manages experimental workflows, and continuously iterates and optimizes directions.
- 3D digital content generation: Generates interactive Three.js websites and digital twin scenes based on precise parameters, and supports GLB model loading, procedural modeling, and multilingual responsive deployment.