Tencent has officially launched and open-sourced its latest large language model, the Hy4 preview, on August 28. The model features a total of 770 billion parameters with 49 billion active parameters and supports a context window exceeding 1 million tokens, showcasing exceptional performance in real-world productivity tasks such as coding, office work, and scientific research.
The Hy4 preview marks a significant expansion in model size, context length, and data scale, with combined advancements in pre-training and post-training delivering another substantial leap in intelligence. Tencent has deeply integrated the model with products like CodeBuddy and WorkBuddy to enhance user experiences in productive scenarios.
In an internal blind test involving 163 experts and 203 engineering tasks, the Hy4 preview scored 2.99 out of 4, slightly outperforming GLM 5.3 (2.92/4) and Kimi K3 (2.94/4). Positioned as a model "born for productivity," the Hy4 preview has benefited from high-quality data co-construction with leading Tencent experts in software engineering, gaming, finance, and security, as well as co-design initiatives with products like WorkBuddy, driving notable improvements across various real-world tasks.
In software engineering scenarios, the model enhances capabilities in understanding, planning, debugging, and validating long-horizon development tasks, while also improving front-end visual aesthetics and interaction quality. In office analytics, it significantly boosts complex office environment comprehension and financial analysis, with a focus on data analysis and cross-file collaboration to streamline workflows from information processing to delivering documents, spreadsheets, and presentations.
For game development, the model strengthens its ability to generate playable prototypes from a single requirement and proficiently use game engines, allowing developers to iteratively refine complex gaming projects through multi-turn interactions. In scientific research, it markedly improves the understanding, reasoning, and problem-solving of complex scientific questions, showing significant progress in AI development, molecular dynamics simulations, condensed matter physics, and fundamental mathematics.
The Hy4 preview is now open-sourced and available across Tencent products including WorkBuddy, CodeBuddy (both domestic and international versions), Yuanbao, and ima, allowing users to experience it directly. Developers can also access the API via Tencent Cloud Tokenhub and OpenRouter. Pricing remains accessible and cost-effective, with input tokens at 6 RMB per million, output tokens at 18 RMB per million, and cache hits at 0.3 RMB per million.
To further gather user feedback and enhance model performance, WorkBuddy and CodeBuddy will offer a two-week limited-time free campaign for users.
Comments