Benchmarking Success
In terms of architecture, Turbo S has adopted the Hybrid-Mamba-Transformer fusion mode – the first time, Tencent says, it has been successfully applied ‘losslessly’ to a very large model.
To demonstrate the model’s speed, the company lists benchmarking for Turbo S against DeepSeek-V3, OpenAI’s ChatGPT 4o, Anthropic’s Claude 3.5 Sonnet and Meta’s Llama 3.1 in areas including knowledge, reasoning, math and code.
Across the 17 sub-categories, it was the overall fastest in 10 (Claude 3.5 Sonnet performed next best, with five ‘wins’), beating ChatGPT 4o in 15 sub-categories and DeepSeek-V3 in 12.
“As the flagship model, Hunyuan Turbo S will become the core foundation of Tencent’s Hunyuan series of derivative models in the future, providing basic capabilities for derivative models such as reasoning, long texts, and codes.” – Tencent
