What's new with FoundationModels this year

This year, Foundation Models got a lot bigger and more flexible. The main updates are:

• A rebuilt on-device model that’s smarter, better at logic, and improved for tool calling.
• Image input support, so you can pass images directly into prompts.
• Server model support, letting apps use frontier or third-party LLMs through the same Swift API.
• New built-in system tools from Vision and Spotlight, including:
  • Barcode reading
  • OCR / structured text extraction
  • A RAG tool powered by Core Spotlight
• Dynamic Profiles, a new primitive for building agentic experiences and composing model behavior.
• Foundation Models Utilities, an open-source Swift package with emerging building blocks for things like:
  • transcript/context management
  • skills
  • ChatCompletion-style server interoperability
• Evaluations Framework, for measuring and validating AI feature quality.
• The FM command line tool, so you can prompt the model from Terminal.
• Python SDK support, bringing the same on-device model to Python.
• Open source foundation models framework, so the Swift APIs can run beyond Apple platforms too.

If you want, I can also break these down by chapter, like models, tools, agentic features, or developer tooling.