Tencent Holdings Ltd. on Friday released and open-sourced Hy4 Preview, a large language model the company said is designed for productivity tasks spanning coding, office work, game development, and scientific research.
The Hy4 Preview model runs on 770 billion total parameters and 49 billion active parameters, with a context window exceeding 1 million tokens, Tencent said in a statement. The model was expanded significantly in size and data volume from its predecessor, with advances in both pre-training and post-training.
Hy4 Preview is available as an open-source model and can be accessed globally through WorkBuddy and CodeBuddy, as well as Yuanbao and ima. Users can connect to the model via API through Tencent Cloud TokenHub and OpenRouter.
The model will be available for free on WorkBuddy and CodeBuddy for two weeks following launch. Free access to the prior Hy3 model on both platforms has been extended until Sept. 30, according to the statement.
Development and test results
In a blind evaluation conducted internally by Tencent involving 163 experts and 203 engineering tasks, Hy4 Preview scored an average of 2.99 out of 4.00, compared with 2.92 for Z.AI Co.'s GLM-5.3 and 2.94 for Moonshot AI's Kimi K3.
The model was developed using training data co-created with Tencent experts across software engineering, gaming, finance, and security, as well as through co-design with products such as WorkBuddy, the company said.
Hy4 Preview contributed to its own development process for the first time, participating in the automated optimization of training methods, data strategies, evaluation frameworks, and low-level operators, Tencent said. The model proposed approaches and iterated based on results, establishing an early-stage recursive self-improvement loop.
The model also autonomously analyzed bottlenecks in its inference system and carried out rounds of optimization on operator fusion and communication. Tencent claims the improvements increased end-to-end throughput by 31.8% compared with baseline, with consistent gains across different context lengths and concurrency levels.
API pricing is set at $0.834 per million input tokens, $2.501 per million output tokens and $0.042 per million tokens for cache hits.