Tencent has released and open-sourced Hy4 preview, a large language model with 770B total parameters and 49B active parameters, featuring a context window exceeding 1 million tokens. According to Tencent, the model is designed for real-world productivity tasks spanning coding, office work, and scientific research.
The model is available as an open-source release and can be accessed globally through Tencent products including WorkBuddy, CodeBuddy, Yuanbao, and ima, as well as via API through Tencent Cloud TokenHub and OpenRouter. Upon launch, Hy4 preview will be available for free on WorkBuddy and CodeBuddy for two weeks. Free access to the previous Hy3 model on both platforms has also been extended until September 30.
According to an internal blind evaluation by Tencent involving 163 experts and 203 engineering tasks, Hy4 preview scored an average of 2.99 out of 4.00, slightly ahead of GLM-5.3 (2.92/4.00) and Kimi K3 (2.94/4.00).
Tencent developed Hy4 preview using high-quality training data co-created with experts across software engineering, gaming, finance, and security domains. In software engineering, the model delivers stronger understanding, planning, debugging, and validation capabilities for long-context development tasks while enhancing visual quality and interaction experience for front-end development. In office productivity, it demonstrates enhanced understanding of complex working environments and improved financial analysis capabilities, with optimization for data analysis and cross-document collaboration.
For game development, Hy4 preview can generate playable prototypes from natural-language requests and work with game engines. In scientific research, the model shows improved capabilities in understanding and solving complex problems, with improvements across AI research, molecular dynamics simulation, condensed-matter physics, and mathematics.
Notably, Hy4 preview contributed to its own development, participating in automated optimization of training methods, data strategies, and evaluation frameworks. The model proposed approaches, ran experiments, and iterated on results, establishing what Tencent describes as an early-stage recursive self-improvement loop. The model also autonomously optimized its inference system, increasing end-to-end throughput by 31.8% compared with baseline through improvements in operator fusion and communication optimization.
API pricing is set at USD 0.834 per million input tokens, USD 2.501 per million output tokens, and USD 0.042 per million tokens for cache hits. Tencent indicates additional models in the Hy4 series are expected to roll out soon.
Key facts
- Hy4 preview has 770B total parameters and 49B active parameters with over 1M token context window
- The model scored 2.99/4.00 in Tencent’s internal blind evaluation, ahead of competing models GLM-5.3 and Kimi K3
- Hy4 preview participated in its own optimization, achieving a 31.8% increase in inference throughput
- Free access available on WorkBuddy and CodeBuddy for two weeks upon launch