Analysis
Tencent has officially entered the competitive arena with a highly impressive release of its new Hunyuan Hy3 preview model. By slashing latency by over 50% and offering incredibly affordable API pricing, they are making advanced Generative AI accessible to everyone. This model showcases fantastic progress in logical inference and complex task execution, proving to be a highly capable and efficient tool for developers.
Key Takeaways
- •The new Hunyuan Hy3 is an Open Source model with 295B total parameters but only activates 21B, maximizing efficiency while supporting a massive 256K context window.
- •Developers will love the incredible speed improvements, with first token latency dropping by 54% and overall response times slashed by nearly half.
- •It features highly competitive pricing on Tencent Cloud, costing as little as 1.2 RMB per million tokens, making it one of the most affordable models in its class.
Reference / Citation
View Original"This model uses a Mixture of Experts architecture that fuses fast and slow thinking, with total parameters of 295B and activated parameters of 21B, supporting a maximum context window length of 256K. Official data shows that the first token latency is reduced by 54%, and the end-to-end duration is reduced by 47%, significantly improving response speed."