Apple Foundation Models Framework 2026: Claude & Gemini

[Alex]: Welcome back to Nerd Level Tech AI Cast, where we turn your caffeine jitters into cutting-edge AI knowledge! I’m Alex, your resident explainer of all things deeply technical.

[Jamie]: And I’m Jamie, here to ask all those “wait, what does that mean?” questions so you don’t have to. Today, we’re unpacking Apple’s Foundation Models Framework 2026, and how it now plays nice with both Claude and Gemini.

[Alex]: That’s right! If you missed WWDC this year—first, how dare you—but second, you missed some seriously big moves from Apple on the AI front. So, let’s break it all down.

[Jamie]: Okay, Alex, you know I have to ask: Foundation Models Framework—what is it, and why are so many devs buzzing about it now?

[Alex]: Great place to start. So, Apple’s Foundation Models Framework is basically a native Swift API that lets developers tap into Apple’s own on-device AI models. It launched last year, but this year, Apple cracked the door wide open—they added support for third-party models like Anthropic’s Claude and Google’s Gemini.

[Jamie]: Wait, so I can use the same code to talk to Apple’s models, Claude, and Gemini? Or do I need to, like, sell my soul to the Swift gods?

[Alex]: You keep your soul—just maybe update your Xcode. Apple introduced a public LanguageModel protocol. Any model that implements this protocol—Apple’s, Claude, Gemini, or even your neighbor’s home-brew AI—can be swapped in and out. So, your app’s logic doesn’t care what’s behind the curtain, it just talks to LanguageModelSession.

[Jamie]: So, it’s like a universal remote for AI models?

[Alex]: Exactly! Instead of juggling remotes for your TV, your soundbar, and your... Roomba, you just have one. Swap out providers by changing a dependency, not rewriting your whole codebase.

[Jamie]: I love it. But tell me: how do Claude and Gemini actually plug in?

[Alex]: Anthropic shipped a Swift package called ClaudeForFoundationModels—open-source, Apache-2.0, but heads up, it’s beta. You drop it in, hand over your Anthropic API key, and you’re good to go. Gemini, on the other hand, comes in via the Firebase Apple SDK. Google calls it “Firebase AI Logic”—it’s a managed service, so you don’t have to wrangle your own backend.

[Jamie]: So, Apple’s on-device, Claude via Anthropic’s package, and Gemini through Firebase. That’s like the AI version of The Avengers, but with less spandex.

[Alex]: And way more Swift code.

[Jamie]: [laughs] Okay, so I’m guessing there’s a catch. Is this all free, or do I need to sell my vintage iPod to pay for API calls?

[Alex]: For small developers, there’s good news. If you’re in Apple’s App Store Small Business Program and your app has fewer than 2 million total first-time downloads, you get access to Apple’s Private Cloud Compute—basically, Apple’s own cloud AI tier—for free. No cloud API costs.

[Jamie]: Nice! So, on-device is free, and if you’re a “small fry” you get cloud power for free, too?

[Alex]: Yep. On-device is always free. Private Cloud Compute is Apple’s server option—it gives you much bigger context windows, like 32,000 tokens compared to the 4,000-ish on-device. So, you can run beefier queries without melting your user’s phone.

[Jamie]: Wait, 32,000 tokens? That’s like... War and Peace in one prompt!

[Alex]: More or less! But these numbers are from session walkthroughs, not official spec sheets, so double-check before you architect your next AI-powered version of Infinite Jest.

[Jamie]: What about new features? Multimodal is the buzzword of 2026—did Apple jump on that train?

[Alex]: Oh, absolutely. Now, you can pass images alongside text, and use on-device Vision tools—like OCR and barcode scanning. There’s also Dynamic Profiles, so you can swap models or tools mid-session, and an Evaluations framework to check your AI isn’t quietly hallucinating that 2+2=5.

[Jamie]: I wish I had Dynamic Profiles for my personality—just swap out “pre-coffee Jamie” for “post-coffee Jamie” on the fly.

[Alex]: [laughs] I think that’s called caffeine. But yes, it’s all about flexibility. And Apple says the whole framework is going open source “later this summer.” So, soon you’ll be able to run the same Swift APIs on your own servers, or wherever Swift runs.

[Jamie]: Wait—open source Apple? Did Tim Cook eat an Android?

[Alex]: [chuckles] Don’t jinx it. The code isn’t live yet—just a promise for now. But they’ve already open-sourced some utilities, so progress is happening.

[Jamie]: Give me a real-world example. How do I actually call Claude from Swift?

[Alex]: Say you want Claude to plan a trip. You import the package, set up your model with your API key, and then: ```swift let model = ClaudeLanguageModel(name: .sonnet4_6, auth: .apiKey("YOUR_KEY")) let session = LanguageModelSession(model: model) let response = try await session.respond(to: "Plan a 4-day trip to Buenos Aires.") print(response.content) ``` Swap “ClaudeLanguageModel” for the default, and you’re back to Apple’s local model. It’s literally a one-line change.

[Jamie]: That’s so much easier than my old days hacking together weird REST calls and praying to the JSON gods.

[Alex]: We’ve all been there. With this, even structured output and streaming responses are handled natively.

[Jamie]: What about Gemini? Anything fancy there?

[Alex]: Gemini hooks in via Firebase, so you get Google’s AI muscle with App Check security—no backend to manage. And, get this—Gemini is now built right into Xcode. So, you can have an AI sidekick help you debug, review, and build features without leaving your IDE.

[Jamie]: So, Xcode is basically turning into Jarvis. Can it make my code less embarrassing?

[Alex]: Only if it can fix my 2 AM “just ship it” commits.

[Jamie]: [laughs] We can all dream.

[Jamie]: Alright, speed round: What are Apple’s third-gen models?

[Alex]: Five models! Two run on device—AFM 3 Core and AFM 3 Core Advanced, which is natively multimodal. Three run in Private Cloud Compute: AFM 3 Cloud for text, ADM 3 Cloud for images, and AFM 3 Cloud Pro for your wildest, most agentic AI needs.

[Jamie]: And are they any good?

[Alex]: Apple’s evaluations say yes—big leaps over last year. But the real test will be in developer hands, not just Apple’s slides.

[Jamie]: And again, open source is coming... “soon.” So, maybe by the time we finally fix that one off-by-one bug, it’ll be here.

[Alex]: [laughs] That’s the dream. [OUTRO]

[Jamie]: Alright, that’s our lightning tour of Apple’s new Foundation Models Framework, Claude, Gemini, and why 2026 is the year everything talks to everything. We’ll post links in the show notes—including that Swift sample, so you can try it yourself.

[Alex]: Thanks for tuning in to Nerd Level Tech AI Cast! If you liked this episode, leave us a review, share it with your favorite developer, and don’t forget to update your dependencies.

[Jamie]: And remember: if your AI starts planning vacations for you, at least make sure it picks somewhere with good Wi-Fi.

[Alex]: See you next time!

[Jamie]: Bye!