Tencent open-sources Hy4 language model with 770 billion parameters
Next-gen AI model Hy4 delivers superior performance in coding, office tasks, and scientific research.
1Massive-scale AI model
Tencent's Hy4 preview boasts 770 billion total parameters and 49 billion active parameters, with a context window exceeding 1 million tokens. The open-sourced model demonstrates exceptional performance on real-world productivity tasks including coding, office automation, and scientific research. Currently available on platforms like WorkBuddy and CodeBuddy, it represents a significant advancement in large language model technology.
2Free access period
Following its launch, Hy4 preview will be available for free on WorkBuddy and CodeBuddy platforms for two weeks. Tencent has also extended free access to Hy3 on these platforms until September 30. Users can experience the model directly through these applications or integrate it via API connections through Tencent Cloud TokenHub and OpenRouter.
3Major generational leap
Hy4 preview shows significant expansion in model size, context length, and data volume compared to previous generations. Improvements in both pre-training and post-training phases resulted in a substantial intelligence leap. In Tencent's blind evaluation involving 163 experts and 203 engineering tasks, Hy4 scored an average of 2.99/4.00, outperforming competitors like GLM-5.3 (2.92) and Kimi K3 (2.94).
4Cross-industry applications
The model enhances understanding, planning, debugging, and validation capabilities for long-context development tasks in software engineering. It shows improved comprehension of complex work environments and financial analysis in office productivity scenarios. For game development, it can generate playable prototypes from single natural language requests, while demonstrating strong problem-solving abilities in scientific research.
5Self-optimization capabilities
Hy4 preview contributed to its own development by participating in automated optimization of training methods and evaluation frameworks. It autonomously analyzed inference system bottlenecks and conducted multiple optimization rounds on areas like operator fusion, achieving 31.8% end-to-end throughput improvement. This demonstrates the model's unique ability to enhance its own infrastructure without human intervention.
6Cost-effective access
Hy4 preview offers affordable access to advanced AI capabilities with API pricing set at $0.834 per million input tokens, $2.501 per million output tokens, and $0.042 per million tokens for cache hits. This pricing strategy aims to make cutting-edge AI technology accessible to wider audiences while maintaining high performance standards across various applications.
7Product integration and feedback loop
The Hunyuan team collaborates deeply with products like CodeBuddy and WorkBuddy to optimize real-world user experience. Using a preview-first approach, the development process incorporates actual user feedback. The model was trained using high-quality data co-created with Tencent experts across software engineering, gaming, finance, and security domains.
8Setting new AI standards
The open-source release of Hy4 preview marks a significant step in intensifying competition within the large language model space. Its superior multi-industry capabilities and autonomous optimization features establish new benchmarks in AI technology. Tencent plans to further refine this technology with official releases following the preview phase, with new models in the Hy4 series expected to launch soon.