Today, Tencent Hunyuan officially announced the comprehensive upgrade of its model matrix, including the flagship fast-thinking model Hunyuan TurboS, the in-depth thinking model Hunyuan T1 upgrade, and the launch of new models based on the TurboS base: the visual deep reasoning model T1-Vision and the end-to-end voice call model Hunyuan Voice.
In addition, the multi-modal models such as Tencent Hunyuan 2.0, Tencent Hunyuan 3D v2.5, and the series of models for game visual generation have also been "updated." This upgrade not only enhances Hunyuan's competitiveness in the AI field but also marks a new progress in Tencent's multi-modal technology.
At the 2025 Tencent Cloud AI Industrial Application Summit, Tao Dao生, Senior Executive Vice President of Tencent Group, stated that the Hunyuan TurboS model has ranked among the top eight globally on the Chatbot Arena evaluation platform, becoming an excellent model second only to DeepSeek domestically. The outstanding performance of TurboS is attributed to its innovative pre-training and post-training technologies, which have improved reasoning abilities in science and mathematics by more than 10%.
The other deep thinking model of Tencent Hunyuan, T1, is also constantly being iterated and recently received a new upgrade. The capabilities of T1 in multiple fields such as competition-level mathematics and common sense question-answering have significantly improved. The score in competition-level mathematics increased by 8%, while the Agent capability for complex tasks improved by 13%. These continuous technological advancements reflect Tencent's commitment and innovative spirit in the AI field.
The newly released T1-Vision model features original long-chain thinking, supporting multi-image input, making the model more efficient in understanding images. Overall performance has improved by 5.3% compared to the previous cascade solution, with a 50% increase in understanding speed.
Meanwhile, the Hunyuan Voice model achieves low-latency voice calls, with a 30% increase in response speed compared to the old version, reducing latency to 1.6 seconds.
In terms of multi-modal generation, Tencent Hunyuan 2.0 leads the industry with its extremely high accuracy and rapid generation capabilities, meeting users' demands for real-time interactive experiences. Additionally, the Hunyuan 3D v2.5 model has made breakthroughs in generation capabilities and controllability, excelling in multiple generation categories.
To better serve various industries, Tencent Hunyuan plans to introduce a large-scale 3D scene generation model capable of free roaming, promoting innovation in gaming and intelligent industries. Meanwhile, Hunyuan is committed to advancing open-source technologies, achieving full-modal openness in image, video, and 3D, and providing free model resources and technical support to partners involved.
Key points:
🔹 Tencent Hunyuan upgrades its model matrix, launching the visual deep reasoning model T1-Vision and the end-to-end voice call model Hunyuan Voice.
🔹 The Hunyuan TurboS ranks among the top eight globally, with significant improvements in scientific reasoning capabilities.
🔹 Tencent Hunyuan is dedicated to open-source technology, supporting continuous iteration and industry application of multi-modal models.