Chinese technology company Tencent has introduced Tencent Hy4 preview, a large language model (LLM) designed to handle coding, office productivity, and scientific research tasks.
The model, now available as open source, contains 770 billion total parameters and 49 billion active parameters, with a context window exceeding one million tokens.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
Tencent has made the model accessible globally via its CodeBuddy and WorkBuddy applications, in addition to other platforms such as Yuanbao and ima. API access is provided through Tencent Cloud TokenHub and OpenRouter.
The company stated that Hy4 preview will be offered free of charge through WorkBuddy and CodeBuddy for the first two weeks following launch.
Free access to the earlier Hy3 model on both platforms will remain available until 30 September.
Hy4 preview features considerable increases in model size, context length, and data volume compared with previous releases.
Tencent reported that developments in both pre-training and post-training phases have led to significant gains in the model’s intelligence.
The company claimed that 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. This result was slightly higher than the scores for GLM-5.3 and Kimi K3, which recorded 2.92 and 2.94 respectively.
It further said that Hy4 preview’s capabilities were developed using high-quality data generated with input from Tencent specialists across software engineering, finance, gaming, and security.
The model has also benefited from continuous collaboration with products such as WorkBuddy.
As a result, Tencent reports enhanced functionality in real-world scenarios. These include improved understanding, debugging, planning, and validation for long-context development in software engineering, as well as enhanced support for office productivity and analytical tasks such as financial analysis, data handling, and document generation.
In game development, the model is said to generate playable prototypes from text commands and supports further development through iterative dialog.
Hy4 preview is also reported to offer improvements for scientific research tasks, including AI research, molecular dynamics simulation, condensed-matter physics, and mathematics.
The company indicated that Hy4 preview contributed to its own creation by suggesting optimisations in training methods and system operations, reportedly establishing a recursive self-improvement loop.
Tencent also stated that the model performed autonomous analysis and optimisation of its inference system, resulting in a throughput increase of 31.8% over the baseline.
Pricing for API access is set at $0.834 per million input tokens, $2.501 per million output tokens, and $0.042 per million tokens for cache hits.
Tencent said it plans to continue its preview-first approach to research and development and expects further models in the Hy4 series to be released in the near future.
In March, Tencent Cloud and ComfyUI partnered to bring Tencent HY 3D Global to ComfyUI’s open-source visual AI platform.
