Tencent Unveils Hunyuan T1 AI Model to Rival DeepSeek in Performance and Price
Tencent Holdings has launched Hunyuan T1, a new artificial intelligence (AI) reasoning model that competes with DeepSeek’s R1 in both performance and pricing.
The Chinese technology giant introduced the model on Friday after a beta run on its chatbot Yuanbao. Hunyuan T1 scored 87.2 points on the Massive Multitask Language Understanding (MMLU) Pro benchmark, surpassing DeepSeek-R1’s 84 points but trailing OpenAI’s o1 model, which scored 89.3.
T1 also performed well in other evaluations, achieving 78.2 points in the American Invitational Mathematics Examination (AIME) 2024, slightly behind R1’s 79.8 and o1’s 79.2. In Chinese language capabilities, T1 matched R1 with 91.8 points in the C-Eval suite evaluation, outperforming o1’s 87.8.
Tencent has positioned T1 as a cost-effective alternative to DeepSeek. The model charges 1 yuan (USD 0.14) per 1 million tokens of input and 4 yuan per million tokens of output. This pricing aligns with R1’s input rate of 1 yuan per million tokens during the day and 0.25 yuan overnight. R1’s output pricing is 16 yuan per million tokens during the day, dropping to 4 yuan overnight.
Tencent claims to be the first in the industry to adopt a hybrid architecture combining Google’s Transformer and Mamba, developed by Carnegie Mellon University and Princeton University. The company states that this approach significantly reduces training and inference costs by lowering memory usage. The model reportedly offers a 200% increase in decoding speed while maintaining the ability to process long text sequences efficiently.
Tech blog NCJRYDS, written by a former JD.com big-data expert, tested T1 and R1 on various tasks. While T1 lost in composing an ancient Chinese poem, it outperformed DeepSeek in interpreting a Chinese word across different contexts. Another blog, GoPlayAI, tested T1 with four mathematical questions, with the model failing the most difficult one after five minutes.
Tencent, which operates WeChat and the world’s largest video gaming business by revenue, is integrating AI as a core revenue stream. The company has incorporated DeepSeek-R1 into its cloud platform and Yuanbao chatbot, offering it alongside its own Hunyuan models.
Tencent Chairman and CEO Pony Ma Huateng recently praised DeepSeek for creating an independent, open-source, and free product. He stated that Tencent follows a “double-core” AI strategy, using both DeepSeek and its own Yuanbao models, similar to its approach in the gaming industry, where it promotes both self-developed titles and those from independent studios.
Tencent launched Hunyuan T1, an AI reasoning model competing with DeepSeek-R1
T1 scored 87.2 on the MMLU Pro benchmark, surpassing R1 but trailing OpenAI’s o1
The model charges 1 yuan per million tokens of input and 4 yuan per million tokens of output
Source: SCMP