Tencent Cloud Launches DeepSeek-V3.1 API
-
On August 22, Tencent Cloud officially announced the launch of the DeepSeek-V3.1 version. This new release not only provides enterprises and developers with more stable and higher-quality services but also lowers the threshold for building intelligent agents through the simultaneous rollout on Tencent Cloud's Agent Development Platform (ADP) and TI Platform, further promoting the widespread application of AI technology.
The release of Tencent Cloud's DeepSeek-V3.1 marks another significant advancement in AI technology within the cloud services sector. Enterprises and developers can now directly call the API interfaces of the new model through Tencent Cloud, enjoying a more efficient and intelligent service experience. The newly launched model not only achieves notable technical improvements but also demonstrates robust performance and flexibility across multiple real-world application scenarios.
Tencent Cloud's Agent Development Platform (ADP) has also integrated the new DeepSeek-V3.1 model. The platform comes equipped with RAG (Retrieval-Augmented Generation) algorithms, workflow capabilities, and intelligent agent development tools, enabling users to quickly build customized agent applications. By importing documents or Q&A pairs, the agent can connect with enterprise multimodal knowledge, delivering stable and accurate knowledge-based responses. Additionally, the platform supports flexible canvas-based workflow orchestration, allowing enterprises to rapidly integrate their system APIs with the agent, ensuring more stable and controllable outputs.
Tencent Cloud's TI Platform has also promptly listed the DeepSeek-V3.1 model, enabling users to quickly deploy customized services based on the TI Platform to meet enterprise-specific needs. The TI Platform is the first in the industry to support enterprise-level fine-tuning and end-to-end inference for the entire DeepSeek model series, offering more efficient and convenient AI model construction and application solutions for industries such as finance, healthcare, manufacturing, and retail. This further reduces the technical barriers and R&D costs for enterprises in adopting large-scale models.
The DeepSeek-V3.1 version has further enhanced tool invocation and agent support, with significantly improved reasoning efficiency. The new version adopts a hybrid inference architecture, allowing a single model to support both reasoning and non-reasoning modes. 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 achieves comparable average performance across tasks while reducing output tokens by 20%-50% compared to R1-0528. Additionally, V3.1 effectively controls output length in non-reasoning mode, maintaining the same model performance with noticeably shorter outputs.
Furthermore, DeepSeek-V3.1 has made significant progress in programming agents. In evaluations such as the SWE (Software Engineering) benchmark and complex tasks in terminal environments (Terminal-Bench), DeepSeek-V3.1 shows clear improvements over previous DeepSeek models. This indicates that the new version is more efficient and accurate in handling complex programming tasks.