Apple's iOS 27 Will Let Users Choose Between Gemini, Claude, and Other AI Models

Apple plans to ship a feature called Extensions in iOS 27, iPadOS 27, and macOS 27 this fall that will let users choose third-party AI models to power their Apple Intelligence features, according to a report from Bloomberg's Mark Gurman published May 5. The feature would allow users to substitute models from Google (Gemini), Anthropic (Claude), and potentially others in place of Apple's default models for tasks including text generation, Writing Tools, Image Playground, and Siri responses. Gurman reports that both Gemini and Claude are already being tested internally. A dedicated App Store section for compatible AI apps is planned to support discovery. The move would represent Apple's most significant concession to third-party AI since the company announced its original OpenAI partnership at WWDC 2024.

What Extensions Would Actually Allow — and What It Wouldn't

The Extensions feature, as described by Bloomberg, operates at the Apple Intelligence integration layer rather than at the OS level. A user with Gemini installed and selected as their preferred model would receive Gemini-powered responses when using Writing Tools to rewrite an email, when asking Siri a knowledge-intensive question, or when generating an image in Image Playground. The underlying Apple Intelligence API calls — the same calls that currently route to Apple's on-device model or to OpenAI's ChatGPT integration — would be redirected to the user's chosen provider.

What Extensions would not change: on-device processing for privacy-sensitive tasks (facial recognition, health data processing, predictive text on-device) would continue to run exclusively on Apple's own neural engine. The Extensions framework applies to the generative AI layer — the tasks that require large language model inference — not to the privacy-first local ML that Apple has built its hardware strategy around. The distinction matters because it means Extensions is an additive feature, not a platform-level restructuring. Apple retains control over which tasks use on-device compute; Extensions governs only the cloud-based generative layer.

Custom Siri voices depending on which model is active are also planned, adding a perceptual differentiation layer: a user running Claude as their default model might hear a distinctly different voice persona when Siri responds, signaling which AI is speaking. This is aesthetically novel but also carries practical implications for user trust and brand differentiation — Anthropic and Google will each have the opportunity to develop distinct voice personalities that compete on tone and style, not just accuracy.

Why Apple Is Opening Up — Regulatory Pressure and Competitive Reality

Apple's decision to introduce multi-model choice is almost certainly not purely voluntary. European regulators have been scrutinizing Apple Intelligence's integration with OpenAI as a potential competition concern, consistent with the Digital Markets Act's interoperability requirements for gatekeepers. The DMA requires that Apple allow third-party services to access core platform functionality on equal terms — and an AI intelligence layer that exclusively routes to Apple's own models or a single partner relationship with OpenAI presents exactly the kind of foreclosure concern regulators have been flagging.

Beyond regulatory pressure, there is a straightforward competitive argument. Android 16 ships with built-in support for Gemini, Claude, and multiple open-source models, including user-selectable defaults. As the AI model ecosystem diversifies rapidly — with Anthropic's Claude 4, Google's Gemini 2.0 Ultra, and a growing field of specialized models all available in 2026 — tying iOS users to a single AI provider creates churn risk among power users who have strong model preferences. Extensions is Apple's answer: retain control of the platform layer while opening the inference layer to a curated set of vetted providers.

The business model implications are meaningful. Apple currently earns a revenue-share arrangement from the OpenAI ChatGPT integration — users who upgrade to paid ChatGPT plans through the Apple Intelligence interface generate App Store commission revenue for Apple. If Extensions allows Google and Anthropic to offer equivalent paid upgrade paths, Apple's AI-related App Store revenue could grow substantially, partially offsetting the cost of maintaining a competitive intelligence layer.

What to Watch

The Extensions announcement, if confirmed at WWDC in June 2026, will trigger immediate negotiation activity between Apple and AI providers not yet in the initial cohort. Watch for whether Meta's Llama-based models and smaller open-source providers are included or excluded — Apple has a history of setting strict quality and privacy thresholds for integration partners. Watch also for the European Commission's formal response: if the DMA compliance officers consider Extensions sufficient to address interoperability concerns, it removes a significant regulatory overhang from Apple's AI roadmap. Finally, the developer extension SDK — which will allow AI companies to build Extensions integrations themselves — is the structural piece: if the SDK is open and well-documented, it can evolve into a competitive ecosystem; if Apple keeps it tightly controlled, Extensions will be a curated list rather than a true marketplace.

Want every AI × Web3 signal the moment it breaks? Subscribe to the BlockAI News daily brief.