Recently, SiJi Flow announced the launch of its latest experimental model, DeepSeek-V3.2-Exp. This version supports a context length of up to 160K and has significantly reduced prices by over 50%. This new version is the result of in-depth optimization based on DeepSeek-V3.1-Terminus, aiming to improve the efficiency of long-text training and inference.
DeepSeek-V3.2-Exp introduces a fine-grained sparse attention mechanism, known as DeepSeek Sparse Attention. According to the SiJi Flow team, the new version performs almost equally to V3.1-Terminus in benchmark tests while significantly improving the ability to handle long texts. This innovative architecture marks an important step forward for the DeepSeek team in advancing model technology.
It is worth noting that since DeepSeek-V3.2-Exp is still in the experimental stage, the stability in certain use cases still needs further verification. To ensure users have a stable experience, SiJi Flow will continue to provide the API service of DeepSeek-V3.1-Terminus.
This release is the 140th model launched on the SiJi Flow platform. Users can experience the powerful capabilities of this new model through the Model Square, as well as various methods such as Claude Code, Gen-CLI, Cline, and Trae.
SiJi Flow is committed to building a trustworthy large model service platform for developers, offering efficient, economical, and reliable API services so that developers can focus on innovation. In addition to DeepSeek-V3.2-Exp, the platform has also launched multiple top-tier models, covering tasks such as language, image, audio, and video, providing developers with a wide range of choices.
Users only need one key to unlock all large models on SiJi Flow. The platform also provides a 14-yuan or 1-dollar trial credit for new users. Additionally, users can use the model performance comparison tool to efficiently evaluate and accurately select the best generative AI solution for their application.
Key Points:
🌟 New Model Release: SiJi Flow launches DeepSeek-V3.2-Exp, supporting a 160K context length.
💰 Significant Price Reduction: The new version's price has dropped by more than 50%, providing users with an economical option.
🔍 Diverse Options: The platform now has 140 models available, covering various application scenarios such as language, images, and audio.
