Playground For LM

What’s New This release introduces a new generation of Playground for LM, bringing local intelligence, multimodal understanding, and smarter model workflows together in one unified experience. On supported devices, Playground for LM can now run compatible MLX models locally, taking advantage of Apple silicon to deliver faster, more private on-device inference. Multimodal workflows have also been redesigned. Native multimodal models can understand images directly, while text-based models can now use a VLM-powered fallback to analyze visual content and convert it into useful text context. This makes image understanding available across more models, even when native vision support is not built in. This update also adds beta support for RAG, allowing models to reference external knowledge and generate more context-aware responses. A smarter routing system now helps choose the right path across local models, vision models, multimodal models, retrieval workflows, and API-based execution. For advanced workflows, the new beta API makes it possible to connect Playground for LM with external tools and use available compute resources, including computer-side processing when configured. These features are currently in beta and may affect stability in some situations. Compatibility, performance, and behavior may vary depending on your device, model, and setup.