iOS 26.3 quietly signals Apple’s openness to deeper third-party AI integration

iOS 26

• iOS 26.3 allows deeper integration for third-party AI apps like Google's Gemini
• Gemini AI Plus subscription offers higher usage limits for $7.99 per month
• NotebookLM now integrates with Gemini on iPhone for enhanced workflows.

Did our AI summary help?

Apple has spent the last year positioning its platform around Apple Intelligence, but a small change in iOS 26.3 suggests the company may be preparing to loosen the reins for third-party AI apps as well. The clearest sign so far is how Google’s Gemini app is evolving on iPhone, with new capabilities that feel far more embedded into the system than before.

This week, Google rolled out expanded functionality for its Gemini app on iOS, including deeper access to NotebookLM directly from the app’s plus menu. While this update is framed as a Google feature, it only works because iOS now allows more flexible contextual inputs and tool switching inside third-party AI apps. That shift is small on paper, but meaningful in direction.

Alongside the iOS-side changes, Google also detailed what users get with its new Gemini AI Plus subscription in the US. Priced at $7.99 per month, AI Plus sits between the free tier and higher-end Pro plans, offering increased daily usage limits without unlocking Google’s most powerful options. On AI Plus, users get up to 90 prompts per day using the Thinking model powered by Gemini 3 Flash, compared to 300 prompts on AI Pro. Gemini 3 Pro is capped at 30 prompts per day on AI Plus, down from 100 prompts on the $19.99 plan.

Context size also scales with the subscription. Free users are limited to 32,000 tokens, roughly equivalent to 50 pages of text. AI Plus increases that to 128,000 tokens, making it far more practical for long documents or sustained conversations. At the top end, AI Pro and Ultra users can access a one million token context window, which Google says can handle up to 1,500 pages of text.

Notably, AI Plus does not increase image generation limits. Nano Banana remains capped at 100 images per day, while Nano Banana Pro stays at three images per day, matching the free tier. Subscribers do get limited access to Veo 3.1 Fast, but Google has not clearly defined how restrictive that access is. Deep Research access is also described as limited and tied more closely to Pro-level plans. Scheduled actions remain capped at 10 at a time, with no differentiation between free and Plus users. These limits also apply to subscribers on the Google One 2 TB plan.

Where iOS 26.3 becomes relevant is in how Gemini now handles NotebookLM on iPhone. The Gemini app for iOS can now pull notebooks directly into prompts while retaining NotebookLM’s native interface style. Users can generate content based on a notebook, attach notebooks to custom Gems, or apply other Gemini tools such as Canvas, Veo, Guided Learning, or Deep Research using the notebook’s contents as context. The same integration is now available to Google Workspace users, extending the feature beyond consumer accounts.

This kind of deep, multi-tool interaction inside a third-party AI app would have been far more constrained on earlier versions of iOS. With iOS 26.3, Apple appears to be allowing richer contextual flows, where apps can act more like AI workspaces rather than single-purpose utilities. That matters because it sets expectations for how AI apps can function on iPhone, even when they are not Apple’s own.

Rather than blocking competitors outright, Apple can keep core system intelligence tightly controlled while still allowing best-in-class third-party tools to flourish within defined boundaries. It also gives Apple leverage, letting users choose alternatives like Gemini without Apple having to build or maintain every specialised AI workflow itself.

The NotebookLM integration is unlikely to be an isolated case. If this pattern continues, iOS updates going forward may increasingly focus on enabling richer AI contexts, deeper app-to-app knowledge sharing, and more flexible tool orchestration, all without explicitly branding these changes as headline AI features.