腾讯混元 Hy4 preview 发布

Tencent Releases Hunyuan Hy4 Preview Model, Solidifying Its Position in the Top Tier of Open-Source Models

IT Home reported on August 28th that Tencent officially released Hy4 Preview today, with a total parameter count of 770B, 49B activated parameters, and a context length of 1M.

According to the official announcement, Hy4 Preview has undergone significant expansion in model size, context length, and data scale. Joint advances in pre-training and post-training have brought another tremendous leap in intelligence, firmly establishing it in the top tier of open-source models.

Through high-quality data collaboration with top experts in Tencent's internal software engineering, gaming, finance, and security domains, Hy4 Preview has achieved remarkable progress across various real-world productivity tasks:

Software Engineering: Enhanced understanding, planning, debugging, and verification capabilities for long-range development tasks, with further improvements in front-end development aesthetics and interactive quality;

Software Engineering: Enhanced understanding, planning, debugging, and verification capabilities for long-range development tasks, with further improvements in front-end development aesthetics and interactive quality;

Office Analytics: Significantly improved complex office environment understanding and financial analysis capabilities, with focused optimization on data analysis and cross-file collaboration, completing the full workflow from information processing to document, spreadsheet, and presentation delivery;

Office Analytics: Significantly improved complex office environment understanding and financial analysis capabilities, with focused optimization on data analysis and cross-file collaboration, completing the full workflow from information processing to document, spreadsheet, and presentation delivery;

Game Development: Enhanced ability to generate playable prototypes directly from requirements, with proficient use of game engines—developers can continuously refine complex game projects through multi-round interactions;

Game Development: Enhanced ability to generate playable prototypes directly from requirements, with proficient use of game engines—developers can continuously refine complex game projects through multi-round interactions;

Scientific Research: Significantly enhanced understanding, reasoning, and problem-solving capabilities for complex scientific research problems, with substantial progress across various scenarios including AI R&D, molecular dynamics simulation, condensed matter physics, and fundamental mathematics.

Scientific Research: Significantly enhanced understanding, reasoning, and problem-solving capabilities for complex scientific research problems, with substantial progress across various scenarios including AI R&D, molecular dynamics simulation, condensed matter physics, and fundamental mathematics.

Hy4 Preview continues deep collaboration with products like CodeBuddy and WorkBuddy, optimizing real user experience in productivity scenarios. According to the official statement, 163 internal experts conducted blind tests on 203 engineering tasks, with results showing Hy4 Preview (average score 2.99/4.00) slightly outperforming GLM-5.3 (average score 2.92/4.00; win 46.8% / tie 12.8% / loss 40.4%) and Kimi K3 (average score 2.94/4.00; win 51.2% / tie 7.9% / loss 40.9%).

Combined with Hyra, Hy4 Preview achieved a major breakthrough on the century-old classic geometry problem—the three-dimensional Blaschke–Lebesgue problem—pushing the volume lower bound from 0.380799 to 0.41104. Compared to the 0.41986 given by the Meissner tetrahedron conjecture, this result brings the conjecture to within just 2% Gap of final proof.

According to the official statement, Hy4 Preview is an early version of the Hy4 iteration, with significant room for improvement in both pre-training and post-training. There are also some known issues, such as tendencies toward over-thinking and excessive self-verification on complex tasks—continuous agile iterations will be pursued.

Similar to Hy3 Preview, we hope to gather extensive real-world feedback through the prompt release of Hy4 Preview, thereby significantly improving the official Hy4 version. At the same time, we will continue to leverage our unique advantage of deep collaboration with Tencent products and experts, continuously enhancing the accessibility and ceiling of productivity.

Hy4 Preview is now open-source, available on Tencent Cloud TokenHub and OpenRouter, and can be experienced on products such as WorkBuddy, CodeBuddy, Yuanbao, and ima.

HuggingFace: https://huggingface.co/tencent/Hy4-preview

HuggingFace: https://huggingface.co/tencent/Hy4-preview

Github: https://github.com/Tencent-Hunyuan/Hy4-preview

Github: https://github.com/Tencent-Hunyuan/Hy4-preview

Modelscope: https://modelscope.cn/models/Tencent-Hunyuan/Hy4-preview

Modelscope: https://modelscope.cn/models/Tencent-Hunyuan/Hy4-preview

Gitcode: https://ai.gitcode.com/tencent_hunyuan/Hy4-preview

Gitcode: https://ai.gitcode.com/tencent_hunyuan/Hy4-preview

Advertising Disclosure: This article contains external redirect links (including but not limited to hyperlinks, QR codes, redemption codes, etc.) for conveying additional information and saving time on screening. Results are for reference only. All articles on IT Home containing external links include this disclosure.

Affiliate Sites: IT Home | ZuiHuihui - Cashback Coupons | iPhone Home | Win7 Home | Win10 Home | Win11 Home

Software Products: IT Home Mobile App | MoFang | ZuiHuihui | YaoZhi