iOS 27 Third-Party AI Models Explained: Gemini, Claude, and More

Apple is reportedly planning to let users choose from competing third-party AI models to power Apple Intelligence features across iOS 27, iPadOS 27, and macOS 27, all expected this fall. Mark Gurman broke the story for Bloomberg yesterday, citing people with knowledge of private plans, as Thurrott reported. If it ships as described, this is not a modest expansion of the existing ChatGPT partnership. It is a structural redesign of what Apple Intelligence is.

Currently, ChatGPT is the only third-party AI integrated into Apple Intelligence, while all native features run on Apple's own in-house models, Thurrott reported today. What's reportedly coming in iOS 27 is a framework that makes model selection a system-level user preference: not a second partnership exception, but a mechanism any qualifying provider could use. Apple is building on a strategy to turn its devices into a thorough AI platform, as 9to5Mac reported, citing Bloomberg's reporting. iOS 27 is set to be formally introduced at WWDC on June 8, per 9to5Mac.

How Apple Intelligence extensions could let you choose an AI model in iOS 27

Apple is calling the mechanism "Extensions." In test versions of iOS 27, the system describes it as letting users "access generative AI capabilities from installed apps on demand, through Apple Intelligence features such as Siri, Writing Tools, Image Playground and more," per Thurrott's account of Gurman's reporting.

The pathway runs through the App Store. AI providers add Extensions support to their existing apps, and once a compatible app is installed, users can set it as their preferred model in Settings, The Verge reported. That single toggle would then route tasks across Siri, Writing Tools, and Image Playground to the chosen model. Google and Anthropic, for example, could each add Extensions support to the Gemini and Claude apps, making both eligible for selection once installed, 9to5Mac noted.

What's new is the scope. Earlier reporting from two months ago covered how Extensions would work with Siri. This week's reports say the framework now extends across Writing Tools and Image Playground as well, 9to5Mac noted. That's the difference between a narrow assistant feature and a platform-wide routing layer.

One consequence worth spelling out: once a user selects a third-party model in Settings, that model handles tasks across all three surfaces. Ask Siri to summarize an email, use Writing Tools to rewrite a paragraph, generate an image in Image Playground the chosen model, not Apple's, does the work. The selection is system-wide, not feature-by-feature, The Verge reported, which means the choice carries real weight. Apple would also let different AI providers use distinct Siri voices, so a response from Claude would sound audibly different from one generated by Apple's own model, per 9to5Mac. Rather than abstracting third-party models behind a unified "Apple Intelligence" surface, Apple would be making the provider's identity audible.

The framework is open in principle, but Apple controls the eligibility criteria, the App Store pathway, and the Settings interface through which users discover and select providers. Openness and curation are not mutually exclusive here. Which providers users actually encounter will depend heavily on how Apple designs those surfaces.

Current reporting leaves several questions unanswered:

• Whether Apple's own models remain the default for all tasks, or whether routing will vary by task type
• How Apple plans to surface data-handling and privacy differences between providers within Settings
• Whether smaller AI providers or open-model apps can participate on equal terms alongside Gemini and Claude

Google's deeper position and the competitive stakes

Apple has already been internally testing integrations with Google and Anthropic, Thurrott reported, citing Gurman's account of two people with direct knowledge. Google's position, notably, appears to go beyond Extensions eligibility. There is a separate reported deal to use Gemini to power native Siri and Apple Intelligence features directly, per 9to5Mac, in addition to whatever role Gemini may play through the user-selectable Extensions framework.

That distinction matters. A user who selects Gemini via Extensions is making an active choice. A user who never touches Settings may still be routed to Gemini under the native integration deal, without knowing it. Those are two different kinds of access to Apple's platform, and only one of them is contingent on the user opting in.

Bloomberg Law argued earlier this year that the Apple-Gemini deal positions Gemini at important access points within the ecosystem, including Siri, drawing a parallel to Apple's search distribution agreement with Google. In the search case, a federal court found that the distribution agreement dampened Apple's incentive to develop its own search capabilities. The Gemini deal raises the same question about AI: if Apple already has access to a foundation model through Google, will it invest in building or partnering with a competing one? Bloomberg Law framed the antitrust concern as slow-moving: the foundation model market currently looks competitive, with ChatGPT, Gemini, and Claude all vying for users, so no clear harm is attributable to the deal today. But default placement at the system level is the kind of structural advantage whose effects tend to show up later.

The Extensions framework, if designed with genuinely open eligibility, could complicate that picture somewhat. Whether it does depends on what Apple hasn't yet disclosed: the specific criteria providers must meet, whether placement within the Settings chooser is neutral across all participating apps, and whether the native Gemini integration confers advantages that Extensions-only access doesn't.

What users would actually notice

The practical setup is straightforward: install a compatible AI app, enable its Extension, select it as the preferred model in Settings, The Verge reported. From that point, the chosen model handles tasks across Siri, Writing Tools, and Image Playground system-wide. The most immediately noticeable signal may be Siri's voice: queries handled by Apple's own system could use one voice, while responses from Claude or ChatGPT could use another, 9to5Mac noted.

Beyond that, the experience gets murkier. What current reporting doesn't address: whether prompts sent to third-party models are processed on-device or transmitted externally, how Apple plans to surface privacy and data-handling differences among providers at the point of selection, and whether premium tiers like Claude Pro or Gemini Advanced would require separate paid subscriptions to unlock full functionality. A user who selects Gemini in Settings may not know, at the moment of selection, whether they're also agreeing to Google's data practices for those requests.

Those details will determine how much practical choice users actually get when they open that Settings screen. A well-designed chooser surfaces the differences. A poorly designed one buries them.

What WWDC needs to answer

The unconfirmed details are where the real story is. Default placement rules which model runs before a user touches Settings matter enormously, because most users won't touch Settings. The full scope of provider eligibility determines whether "open" means a handful of well-capitalized partners or a genuinely competitive field that includes smaller models. Privacy disclosure design will dictate whether users can make an informed choice or just a nominal one.

Those specifics will determine whether iOS 27's approach to third-party AI models represents genuine user choice or a more controlled form of curation dressed as openness. WWDC opens June 8. That's when reported plans meet announced reality.