New open-source models have been added to the Foundation Models Catalog: solutions from IBM (Granite), Alibaba (Qwen), DeepSeek, Microsoft (Phi), Mistral AI, and OpenAI.
In the updated version, users can test, launch, and integrate large language models (LLMs) into applications and corporate processes. The platform provides monitoring and automation tools, improving model performance.
Among the key updates is the deployment of AI models on dedicated infrastructure with autoscaling capabilities and access through private networks to enhance stability and information security. Observability tools have been added, allowing control over the state and performance of models, including Inference Server logs and metrics.
The platform architecture has been updated: vLLM, a high-performance open-source framework for running LLMs, is used as the main component for inference. New open-source models from IBM, Alibaba, Microsoft, and other companies have been added to the catalog. Service management is available via REST API, which simplifies integration into business processes.
The updates also affected the user interface: filtering by characteristics has been introduced, making it easier to search for and select suitable models.
