On August 22, Tencent Cloud officially launched the DeepSeek-V3.1 version. This new version not only provides enterprises and developers with more stable and high-quality services, but also synchronizes with the Tencent Cloud Intelligent Agent Development Platform (ADP) and TI platform, further lowering the barrier to building intelligent agents and promoting the widespread application of AI technology.
The release of Tencent Cloud's DeepSeek-V3.1 version marks another important advancement in AI technology within the cloud service field. Enterprises and developers can now directly call the API interface of the new model through Tencent Cloud, enjoying a more efficient and intelligent service experience. The newly launched model has achieved significant technical improvements and demonstrates strong performance and flexibility in multiple practical application scenarios.
The Tencent Cloud Intelligent Agent Development Platform (ADP) has also integrated the new DeepSeek-V3.1 model. The platform is equipped with RAG (Retrieval-Augmented Generation) algorithm, workflows, and intelligent agent development capabilities, allowing users to quickly build custom intelligent agent applications. By importing documents or Q&A pairs, the intelligent agent can connect to enterprise multimodal knowledge, achieving stable and accurate knowledge QA results. In addition, the platform supports canvas-style flexible workflow arrangement, enabling quick integration of enterprise system APIs into the intelligent agent, ensuring more stable and controllable output from the intelligent agent.
Tencent Cloud's TI platform has also immediately listed the DeepSeek-V3.1 model, allowing users to quickly deploy customized services based on the TI platform to meet personalized needs of enterprises. The TI platform is the first in the industry to support enterprise-level fine-tuning and inference full-chain functions for the entire DeepSeek series of models, providing more efficient and convenient AI model construction and application solutions for multiple industries such as finance, healthcare, manufacturing, and retail, further reducing the technical barriers and R&D costs for enterprises in large model applications.
The DeepSeek-V3.1 version has further enhanced tool calling and intelligent agent support, with a significant improvement in thinking efficiency. The new version adopts a hybrid reasoning architecture, supporting both thinking mode and non-thinking mode with a single model. Compared to the previous DeepSeek-R1-0528 version, DeepSeek-V3.1-Think can provide answers in a shorter time. Test results show that after chain-of-thought compression training, V3.1-Think maintains comparable average performance to R1-0528 while reducing the number of output tokens by 20%-50%. At the same time, the output length of V3.1 in non-thinking mode is effectively controlled, maintaining the same model performance even with significantly reduced output length.
In addition, DeepSeek-V3.1 has made significant progress in programming intelligent agents. In the code repair evaluation SWE and complex tasks in the command line terminal environment (Terminal-Bench), DeepSeek-V3.1 shows significant improvements compared to previous DeepSeek series models. This indicates that the new version is more efficient and accurate when handling complex programming tasks.
