iOS 27 Will Let You Kick ChatGPT Off Your iPhone and Replace It With Claude or Gemini

What Extensions Actually Does

The Bloomberg report by Mark Gurman, citing people with knowledge of Apple's plans, describes a system that goes far beyond the current ChatGPT integration in iOS 18 and 19. Under the current setup, Siri can optionally route certain queries to ChatGPT, but the integration is invisible to most users and the routing logic is controlled entirely by Apple.

Extensions changes the architecture. Users will be able to open a new settings panel -- likely within the Apple Intelligence & Siri section -- and select a default AI provider from a list that will initially include OpenAI's ChatGPT, Google's Gemini, and Anthropic's Claude. The selection applies system-wide: every Apple Intelligence feature that currently calls an external model will honor the user's choice.

The framework is reportedly designed to accommodate additional providers over time. Apple has already signed a multi-year, $1 billion-per-year deal with Google to integrate Gemini into Siri and Apple Intelligence, announced in January 2026. That deal established the precedent for third-party AI on Apple platforms. Extensions generalizes it.

For developers, the implications are significant. If Apple exposes an API for AI model integration -- and the "Extensions" naming strongly suggests it will -- any AI company that builds a compliant integration could theoretically appear in the settings menu. This would turn Apple Intelligence from a product into a platform, with Apple collecting its standard 15–30% commission on any paid AI subscriptions routed through the system.

Why Apple Is Doing This -- And Why It Has No Choice

The official framing will emphasize user choice. The reality is more structural: Apple cannot build a best-in-class AI model, and it has stopped pretending it can.

The evidence has been accumulating for two years. Apple's personalized Siri features, announced at WWDC 2024, were repeatedly delayed -- first from iOS 18.4 to "2025," then to "spring 2026," with an internal target of iOS 26.4. The features still have not shipped. Apple recently agreed to pay $250 million to settle a class-action lawsuit over false advertising of those same AI capabilities. The Verge noted that Extensions "could represent Apple's most significant concession that its in-house AI development has fallen behind the frontier labs."

Meanwhile, OpenAI ships GPT-5.5. Anthropic ships Claude Opus 4.7. Google ships Gemini 3.1 Ultra. DeepSeek ships V4 with open weights at 1/30th the inference cost. The AI model landscape moves on a monthly cadence. Apple's development cycle -- annual OS releases with features locked in months before WWDC -- is structurally incompatible with an industry where the state of the art changes every six weeks.

Extensions is Apple's answer to this structural mismatch. If Apple cannot build the best model, it can build the best platform for models to compete on. This is not a retreat. It is a pivot from competing on AI capability to competing on AI distribution -- and distribution is a game Apple has never lost.

The strategy mirrors what Apple did with default apps. For years, Safari, Apple Mail, and Apple Maps were the only options on iOS. Regulators and user demand eventually forced Apple to allow third-party defaults for browsers, email, and navigation. The result was not a mass exodus from Apple's apps -- most users never changed the defaults -- but the option to change them eliminated a major criticism of the platform. Extensions applies the same logic to AI: most users will likely keep whatever Apple sets as the default, but the existence of choice neutralizes the argument that Apple is locking users into inferior AI.

OpenAI Loses Its Crown Jewel

No company has more to lose from Extensions than OpenAI. The ChatGPT integration in Apple Intelligence, which arrived with iOS 18.2 in December 2024, was one of the most valuable distribution deals in AI history. It put ChatGPT in front of hundreds of millions of iPhone users as the default -- and only -- third-party AI option on the world's most lucrative mobile platform.

That exclusivity is about to evaporate.

OpenAI built its consumer brand partly on being "the AI that comes with your iPhone." When iOS 27 launches, ChatGPT will be one option among several. Users who prefer Anthropic's writing style, Google's search integration, or simply distrust OpenAI's data practices will be able to switch with a single settings change. Every user who switches is a user OpenAI acquired through Apple's distribution and will lose through Apple's platform neutrality.

The timing is especially awkward for OpenAI, which is preparing for what could be the largest technology IPO in history -- a listing reportedly targeting a $1 trillion valuation. Losing platform exclusivity on the world's most valuable consumer device ecosystem, weeks before an IPO roadshow, is the kind of narrative risk that makes underwriters nervous.

Google and Anthropic, by contrast, gain something neither could have built independently: a direct channel to hundreds of millions of high-intent users who are already comfortable asking their phone for AI assistance. For Anthropic, which has historically lagged in consumer distribution compared to OpenAI and Google, appearing as a first-class option in iOS Settings is arguably more valuable than any marketing campaign it could run.

The $1 Billion Google Deal Was the Trojan Horse

The Extensions strategy did not emerge from nowhere. In January 2026, Apple and Google announced a partnership that, at the time, seemed like a straightforward outsourcing deal: Google's Gemini models would power the next generation of Siri and Apple Intelligence, reportedly at a cost of $1 billion per year to Google.

That framing missed the larger strategic significance. By signing a deal with Google -- a direct competitor to OpenAI -- Apple established the legal, technical, and contractual foundation for multi-model support. Once Gemini was running inside Apple Intelligence, excluding Claude became harder to justify. Once Claude was in, the framework for arbitrary third-party models was the logical next step.

The Google deal was not a replacement for OpenAI. It was a proof of concept for a multi-provider architecture. Apple built the plumbing for one third-party model and realized the same plumbing could support any number of them. Extensions is that realization turned into a product feature.

The $1 billion figure is also worth examining in reverse. If Google is paying Apple $1 billion per year to be the default AI on iPhones, and Apple is now opening that default slot to competition, the annual value of iPhone AI distribution is at minimum $1 billion -- and likely much higher once multiple providers bid for prominence. Apple may have discovered that selling AI distribution rights is a better business than building AI itself.

The Walled Garden Was Never the Product

The iPhone's closed ecosystem was always a means to an end, not the end itself. Apple's goal was never to restrict user choice for its own sake. The goal was control -- over quality, over privacy, over the user experience, and over the revenue streams that flow through the platform. If opening the platform to third-party AI models serves those goals better than keeping it closed, Apple will open it.

The EU's Digital Markets Act has been forcing Apple's hand on interoperability for years, mandating third-party app stores, browser engines, and device interoperability. But the DMA doesn't require Apple to let users choose their AI model. Extensions goes beyond any regulatory requirement. It is a voluntary strategic move, not a compliance exercise.

The distinction matters because it reveals Apple's actual calculus. The company is not opening up because regulators made it. It is opening up because the economics of being the AI platform for 2.2 billion devices are better than the economics of being one AI developer among many -- especially when your own AI development is behind schedule.

The walled garden strategy worked when Apple could credibly claim that its in-house technology was best-in-class. The iPhone's camera, the Mac's build quality, the Watch's health sensors -- these were areas where Apple's vertical integration produced genuinely superior products. AI is different. The best models are being built by companies whose entire business is building models, iterating on a weekly cadence that Apple's annual release cycle cannot match. In this environment, being the platform where the best models compete is a stronger position than being the company that ships the seventh-best model on a predictable schedule.

The Parallel to Default Apps -- And Why This Is Bigger

Apple has opened default app slots before. Starting with iOS 14 in 2020, users could choose third-party browsers and email clients as defaults. iOS 18.2 added default messaging and calling apps in the EU. Each opening followed the same pattern: regulators demanded it, Apple resisted, then Apple complied -- and almost nothing changed because most users never touched the settings.

Extensions is different in three ways that make its impact potentially much larger.

First, AI model quality varies dramatically in ways that browsers and email clients do not. Safari and Chrome render the same web pages. Apple Mail and Gmail send the same emails. But Claude, Gemini, and ChatGPT produce fundamentally different outputs for the same prompt. A user who switches from ChatGPT to Claude because they prefer Anthropic's writing style gets a materially different product, not just a different interface. The incentive to switch is real.

Second, AI is currently a premium subscription product, not a free bundled feature. ChatGPT Plus costs $20 per month. Claude Max costs $100 per month. If users are paying for an AI subscription, they will want that subscription to work everywhere -- including inside Siri and Writing Tools. The integration point that Extensions provides is not a nice-to-have. For paying subscribers, it is the reason to subscribe to one service over another.

Third, the switching cost is near zero. Moving from Safari to Chrome requires importing bookmarks, learning a new interface, and trusting a different company with your browsing history. Moving from ChatGPT to Claude inside Apple Intelligence requires tapping a different name in a settings menu. The default is sticky, but the friction to change it is lower than any platform default switch in Apple's history.

The comparison to default browser choice is instructive but misleading. Browser choice was a compliance exercise that changed little. AI model choice is a product decision that could reshape which AI companies win consumer markets -- because the distribution is that valuable and the switching cost is that low.

What Happens at WWDC 2026

WWDC 2026, expected in early June, will be the moment Apple has to sell this strategy to developers, users, and -- perhaps most importantly -- Wall Street. MacRumors confirmed that the Extensions feature will be a "key tentpole" of the keynote, with Apple planning dedicated sessions for developers on the API integration framework.

The developer story is straightforward: an API that lets any AI company integrate with Apple Intelligence, with Apple handling distribution, privacy, and payment processing. Developers get access to Apple's user base. Apple gets its platform commission. Users get choice. The narrative is clean.

The user story is trickier. Apple has spent 15 years training iPhone owners not to think about which browser engine or email client they're using. Adding an AI model selector to Settings introduces a decision that most users are not equipped to make. The difference between Claude and Gemini is not as intuitive as the difference between Safari and Chrome. Apple will need to build comparison interfaces, explainer screens, and smart defaults that guide users toward sensible choices without overwhelming them.

The Wall Street story depends on whether Apple can monetize the platform. A 30% commission on AI subscriptions routed through iOS -- the same rate Apple charges for App Store purchases -- could generate significant revenue if AI subscriptions become as common as streaming services. But AI companies may resist the Apple tax, and regulators may view a 30% cut on AI model access as anticompetitive. The DMA's interoperability requirements could complicate Apple's ability to charge for distribution that the EU mandates.

The Industry Rearranges Around Apple's Decision

Extensions will force every AI company to answer a question they have not had to seriously consider: what is your iOS strategy?

OpenAI has one, and it is about to become obsolete. Anthropic and Google now have one handed to them. But the implications extend further. Meta, which has been building Llama models and integrating them into WhatsApp and Instagram, may need to decide whether to join Apple's platform or compete with it. Microsoft, whose Copilot strategy depends partly on being the default AI in Office and Windows, may face a similar choice if Apple's platform model proves successful.

The startups building specialized AI -- Perplexity for search, Midjourney for images, Runway for video -- face the most interesting calculus. Extensions, if it supports domain-specific AI providers, could turn iOS into a marketplace where users pick a writing AI, a coding AI, and an image AI independently. This would fragment the market in ways that favor specialized players over general-purpose chatbots.

For Apple, the risk is not that Extensions fails. It is that Extensions succeeds too well -- that users flock to third-party AI, Apple's own models become irrelevant, and the company's AI strategy reduces to collecting platform rent. That outcome is profitable in the short term but leaves Apple dependent on suppliers for its most important software feature. For a company that still describes itself as a product company first, that dependence may prove uncomfortable even if the checks clear.

iOS 27's Extensions framework is something genuinely rare in technology: a platform owner voluntarily ceding control of its most strategically important feature layer to competitors. Apple is not doing this because it wants to. It is doing this because building AI at the frontier is harder than building the platform that distributes it -- and the platform business, for Apple, has always been the better business anyway. Tools like remio that sit at the intersection of AI models and knowledge management face the same question: integrate with the best models or build your own. The answer, increasingly, is integrate.

Frequently Asked Questions

When will iOS 27 be released?

iOS 27 is expected to be announced at WWDC 2026 in June, with a public release in September 2026 alongside new iPhone hardware.

Can I use Claude instead of ChatGPT on my iPhone right now?

Not as a system-wide default. Currently, ChatGPT is the only third-party AI integrated into Apple Intelligence. iOS 27's Extensions framework will allow users to choose Claude, Gemini, or other providers.

Will Apple charge for third-party AI model access?

Apple has not confirmed pricing, but if Extensions follows the App Store model, Apple may take a 15-30% commission on AI subscriptions routed through iOS. Providers may also pay for default placement, similar to Google's reported $1 billion annual deal.

Does this mean Siri is being replaced?

No. Siri remains Apple's voice assistant. Extensions changes which AI model handles the intelligence behind Apple Intelligence features like Writing Tools and Image Playground. Siri is the interface; third-party models are one of the engines.

The company that once told users they were holding their phone wrong is about to tell them they can choose whichever AI they want. That shift says less about Apple's philosophy than about the structural reality of AI development in 2026: no single company, no matter how wealthy, can keep up with the entire field. The smartest play is not to try -- it's to own the field where everyone else competes.