Mark Gurman: iOS 27 lets model swap

Apple’s AI problem has never just been model quality. It’s product shape. Right now Apple Intelligence feels like a bundle of features with one branded front door and one obvious outside partner — ChatGPT. The new twist is that Apple may stop pretending one model fits every job. Mark Gurman says iOS 27, iPadOS 27, and macOS 27 are being built to let users pick third-party AI models for system features like text and image generation this fall. (bloomberg.com) ### What actually changed? The important shift is not “Apple will use more AI.” That part was obvious. The shift is that Apple may expose model choice to users as a product setting. Gurman’s report says outside services would be selectable across Apple software, instead of Apple quietly deciding when to hand a request to one partner behind the scenes. (bloomberg.com) ### Why is that a big deal? Because it changes Apple Intelligence from a single assistant into a broker. A broker decides which model handles which task, keeps the interface consistent, and tries to hide the mess underneath. That is a very diffe(bloomberg.com)and trust with users, but rivals still move faster on frontier models. (bloomberg.com) ### Which models are in play? The names showing up around this report are the obvious ones — OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude. MacRumors, summarizing the Bloomberg report, says Apple has already signed a deal with Google(bloomberg.com) to rival assistants beyond ChatGPT. (macrumors.com) ### Where would users feel it first? The clearest early surfaces are Writing Tools and Image Playground. Those are the easiest places to swap models because users already think of them as generation features — rewrite this, summarize that, make an image in this style. A deeper Siri handoff(macrumors.com) context. (bloomberg.com) ### Why not just keep ChatGPT? Because different models are good at different things. Claude often gets picked for careful writing and long-context reasoning. Gemini is tightly tied to Google’s broader AI stack. ChatGPT still has the strongest c(bloomberg.com)ic, applied to foundation models. The catch is that models are not interchangeable widgets — they have different safety rules, latency, pricing, and failure modes. (bloomberg.com) ### What gets harder for Apple? Governance. If a user picks Claude for writing, Gemini for images, and something else for Siri later, Apple has to explain what data goes where, when a request leaves the device, which provider answered, and what (bloomberg.com)fer a sane fallback. That is not a side detail — it is the product. This last part is an inference from the reported architecture, but it follows directly from Apple turning model choice into a user feature. (bloomberg.com) ### Why now? Because Apple is still catching up after a slow AI rollout, and WWDC on June 8 is the obvious place to reset the story. Gurman has been reporting for months that iOS 27 is the big Siri and Apple Intelligence rebuild, with deeper cam(bloomberg.com) stopped trying to win the model race head-on and is trying to own the interface instead. (bloomberg.com) ### Bottom line If this ships, Apple Intelligence becomes less like one assistant and more like a control panel for competing AIs. That could be the smartest move Apple has in AI — but only if the routing, privacy, and fallback behavior feel invisible when things go right and very clear when they do not. (bloomberg.com)