Skip to main content

Release Note v2025.03

1. Public Inference Service Now Supports API Access​

  • Users can now access the public inference service via API, enabling seamless integration into applications and workflows for more convenient inference calls.

2. kTransformer Support Added to the Inference Engine​

  • The inference engine now supports kTransformer, offering an ultra-low-cost solution for large model inference. This allows AI models to run efficiently on low-end hardware without relying on expensive high-end GPUs.

3. Supports dynamic modification of inference framework parameters​

  • Supports dynamic modification of inference framework parameters, allowing users to customize parameters when creating a dedicated instance to meet specific needs.

4. New Widget Support in Dedicated Instance Sandbox​

  • Image-text-to-text: Converts image and text inputs into text, ideal for multimodal tasks.
  • Feature extraction: Supports feature extraction, enhancing model analysis and vectorization tasks.

5. MCP Server Support in Application Space​

  • Users can now deploy dedicated MCP Servers directly within the application space.

6. Model Source ID Display Added​

  • Model details page now includes a model source ID, improving traceability and transparency of model origins.

7. Enhanced Interaction for Events & Blogs​

  • New tags, search, and comments features make it easier to find relevant content and engage with the community.

8. AgenticOps Product Map Released​

  • We introduce AgenticOps, an innovative framework for Agent development, deployment, and optimization. Our product suite—CSGHub, StarShip, and AutoHub—is closely integrated with AgenticOps, forming a complete ecosystem for Agent development and operations. A detailed product map is now available on the homepage.

Version List​

Click Version List to check all release notes.