Tencent has released and open-sourced Tencent Hy4 preview, a large language model designed for coding, office productivity, game development, and scientific research.
The model packs 770 billion total parameters, with 49 billion active parameters, and supports a context window exceeding 1 million tokens. Tencent says the combination of larger model size, expanded training data, and improvements to pre- and post-training puts Hy4 preview among the top tier of open-source models.
Hy4 preview is available globally through Tencent’s WorkBuddy and CodeBuddy, as well as Yuanbao and ima. Developers can also access it through APIs via Tencent Cloud TokenHub and OpenRouter.
Hy4 targets real-world productivity
Rather than focusing solely on general chatbot capabilities, Tencent developed Hy4 preview around practical workflows across multiple industries.
The company says training data was developed with Tencent experts in areas including software engineering, gaming, finance, and security, alongside deeper integration with products such as WorkBuddy.
In an internal blind evaluation involving 163 experts and 203 engineering tasks, Hy4 preview achieved an average score of 2.99 out of 4.00. Tencent reported scores of 2.92 for GLM-5.3 and 2.94 for Kimi K3 in the same evaluation.
The results suggest Tencent is positioning Hy4 as a model intended to compete on practical task performance rather than parameter count alone.
Coding, office work, and game development
Hy4 preview is designed to handle long-context software engineering tasks, with improvements in code understanding, planning, debugging, and validation.
The model also targets front-end development, where Tencent says it can improve both the visual quality and interaction experience of generated interfaces.
For office and analytical workflows, Hy4 is designed to process information across multiple documents and assist with data analysis. Tencent says the model can support workflows ranging from information processing to creating documents, spreadsheets, and presentations.
Game developers can also use natural-language prompts to generate playable prototypes and interact with game engines. Complex projects can then be refined through multiple rounds of conversation.
A model that helped optimize itself
One of Hy4 preview’s more unusual claims is its involvement in its own development process.
Tencent says the model participated in automated optimization of training methods, data strategies, evaluation frameworks, and low-level operators. It could propose approaches, run experiments, assess results, and feed the resulting code, logs, and feedback into subsequent iterations.
Tencent describes this as an early-stage recursive self-improvement loop.
Hy4 also analyzed bottlenecks in its own inference system and carried out optimization involving areas such as operator fusion and communication.
According to Tencent, these changes increased end-to-end inference throughput by 31.8% compared with the baseline, with improvements maintained across different context lengths and concurrency levels.
Hy4 extends into scientific research
The model is also being positioned for research-heavy workloads.
Tencent says Hy4 preview has improved its ability to understand and reason through complex scientific problems, with reported gains in areas including AI research and development, molecular dynamics simulation, condensed-matter physics, and fundamental mathematics.
Its more than 1 million-token context window is particularly relevant to workloads involving large amounts of technical information, source material, or code.
Open-source model comes with low-cost API access
Tencent is making Hy4 preview available as an open-source model while also offering access through its own applications and third-party platforms.
For API users, pricing is set at US$0.834 per million input tokens, US$2.501 per million output tokens, and US$0.042 per million tokens for cache hits.
WorkBuddy and CodeBuddy will offer Hy4 preview for free for two weeks following launch. Tencent has also extended free access to Hy3 on both platforms until September 30. Tencent says the next batch of models in the Hy4 series is expected to arrive soon, continuing its preview-first approach of using real-world feedback to refine its models before wider releases.
With its combination of open-source availability, a 1-million-plus-token context window, and an emphasis on practical workloads, Hy4 preview marks Tencent’s latest attempt to push its Hunyuan models deeper into the increasingly competitive open-source AI market.
