iOS 27 to Let Users Choose AI Models by Task

According to TechCrunch on May 5, Apple is preparing a significant shift in how iOS manages artificial intelligence tasks: instead of forcing a single model for all functions, iOS 27 would let users select which third-party model they want to use based on the task. The news comes as Apple Intelligence continues to disappoint a significant portion of its most demanding user base.

The current model, Apple Intelligence as a single layer with occasional access to ChatGPT, has proven insufficient for those who need more advanced capabilities or simply prefer other providers. The described change points toward a task-based selection architecture: one model for writing, another for search, another for code, and so on.

What this would mean in practice

If the information is confirmed, iOS 27 would become the first mass-market mobile platform with a model routing system directly exposed to end users. This has several concrete implications:

• For users: choosing between models based on their preferences for privacy, cost, or performance, without leaving the operating system.
• For AI providers: direct access to hundreds of millions of devices through an official channel, under the terms Apple imposes.
• For Anthropic in particular: Claude, in any of its current variants—Opus 4.7, Sonnet 4.6, or Haiku 4.5—could appear as a selectable option within the system, something that today is only possible through third-party apps or Claude's own app.

Why this matters beyond the headline

Apple's move is not altruistic. Opening the system to external models transfers part of the reputational risk: if a third-party model gives a problematic response, responsibility becomes diffused. At the same time, Apple maintains control of the presentation layer, permissions, and presumably the commercial agreements with each provider.

This echoes what happened with browsers on iOS: Apple took years to allow alternative rendering engines, and when it did, pressured by regulation in Europe, it did so under strict conditions. A similar pattern should be expected here: controlled openness, with Apple as the arbiter of which models enter and under what terms.

From the Claude ecosystem side, the relevant question is whether Anthropic will negotiate a first-class integration or remain as a secondary option. Opus 4.7's 1M token context window or Haiku 4.5's speed could be solid commercial arguments depending on the type of task Apple assigns to each category.

For whom this changes anything today

For now, nothing. iOS 27 has not been released and the news is based on sources not officially confirmed by Apple. For developers working with Claude integrations, whether via MCP servers, Claude Code, or the direct API, this hypothetical scenario does not alter any current workflows.

It does make sense, however, for product teams with apps in the Apple ecosystem to start thinking about how they would position their Claude integrations if the model selection system materializes: as the default model for a specific category? As an advanced option for users with a subscription? The architectural decisions made now on mid-term projects could be affected.

---

Editorial perspective: The idea has technical merit, model routing by task is already done manually in enterprise environments, but Apple's history with genuine platform openness calls for caution. We'll see how much of this makes it to WWDC and in what form.