Tencent Unveils Hunyuan Turbo S AI Model

Tencent has released a new artificial intelligence model named Hunyuan Turbo S, which the company claims is faster than the DeepSeek-R1 model. The model is designed to provide near-instant replies by doubling the output speed and reducing the delay of the first word by almost 44%. This new model combines a short thinking chain for immediate responses with a slow-thinking chain for reasoning capabilities, making it comparable to leading models like DeepSeek-V3 and OpenAI's GPT-4o in benchmarks for math, reasoning, and knowledge.

The Hunyuan Turbo S model employs a Hybrid-Mamba-Transformer fusion to reduce computational complexity, allowing it to handle long sequences effectively while maintaining the ability to understand complex ideas. Tencent stated that this is the first successful application of the Mamba architecture to an ultra-large Mixture of Experts model without damage. This architecture also significantly reduces training and deployment costs, positioning Turbo S as a core model for future AI applications in inference, text, and code generation.