Tencent Launches Hunyuan T1 Reasoning Model to Compete with DeepSeek and OpenAI
Chinese tech giant Tencent has launched its new Hunyuan Thinker-1 (T1) reasoning model, offering advanced data processing capabilities comparable to DeepSeek’s R1.
Key Points
T1 is built to maintain clear content logic and produce clean, organized text with a remarkably low hallucination rate, a key measure of AI accuracy and reliability.
T1 scored 87.2 on the MMLU Pro benchmark, surpassing DeepSeek-R1’s 84 but trailing OpenAI’s o1 at 89.3. It also scored 78.2 in AIME 2024 and excelled in Chinese language with 91.8 in C-Eval, matching R1 and beating o1’s 87.8.
T1 charges 1 yuan per million input tokens and 4 yuan per million output tokens, matching R1’s input rate but offering lower output costs than R1’s daytime rate of 16 yuan, which drops to 4 yuan overnight.
Tencent claims T1 is the first to use a hybrid architecture combining Google’s Transformer and Mamba, developed by Carnegie Mellon and Princeton University.
After a successful preview, T1 is now accessible on various platforms, including Tencent’s AI assistant Yuanbao.
Background
Tencent, the operator of WeChat and the world’s largest video gaming business by revenue, is making AI a key focus for generating new revenue.
Tencent is making AI a key revenue stream, integrating DeepSeek-R1 into its cloud platform and Yuanbao chatbot alongside its own Hunyuan models.
Tencent CEO praised DeepSeek’s open-source approach, adopting a “double-core” AI strategy similar to its gaming industry model of supporting both in-house and independent products.
News Gist
Tencent unveiled the Hunyuan T1 reasoning model, offering advanced data processing with low hallucination rates.
It outperformed DeepSeek-R1 in benchmarks like MMLU Pro and C-Eval while using a hybrid Transformer-Mamba architecture.
T1 is accessible via Yuanbao and other platforms.