Tencent says its new 'T1' reasoning model can outperform DeepSeek
What's the story
Chinese tech giant, Tencent, has officially launched its T1 reasoning model. The new model is said to offer quicker response times and improved capabilities for processing extended text documents. According to Tencent's WeChat account, the T1 can "keep the content logic clear and the text neat and clean," while maintaining an "extremely low" hallucination rate. This launch comes amid increased competition in China's artificial intelligence (AI) sector.
AI competition
T1 model to rival DeepSeek's R1
The T1 model's release comes after DeepSeek introduced models with similar or better performance at cheaper rates than Western systems. Tencent had earlier released a preview version of the T1 via its AI assistant app, Yuanbao. The official version will be powered by Tencent's Turbo S foundational language model, which was unveiled last month and is said by the company to process queries faster than rival DeepSeek's R1 model.
Performance comparison
Outperforming DeepSeek in knowledge benchmarks
A chart published by Tencent comparing the T1 model to DeepSeek R1, revealed that the former outperformed the latter on some knowledge and reasoning benchmarks. This performance edge could give Tencent a leg up in China's rapidly evolving AI landscape. The company has notably ramped up its AI investments in recent months and announced plans to further boost capital expenditure in 2025 after aggressive spending throughout 2024.