Apple shipped the new Siri, and Google's model is doing the talking

Apple opened the English public beta of its rebuilt Siri on 15 September, the first time the overhauled assistant has been available to anyone outside the developer programme. The models underneath it were built with Google and based on Gemini. Apple has been fairly open about that arrangement rather than leaving it to teardowns and speculation.

What actually changed in the Gemini powered Siri

Siri now has its own app. That sounds like housekeeping, but it changes the shape of the thing. The old Siri was an interruption you summoned and hoped for the best with. A standalone app puts it in the same posture as ChatGPT or Gemini, where the conversation is a place you go rather than a moment you trigger.

The second change is memory. The phase two rollout handles more than twenty consecutive dialogue turns, up from the much shorter window in the earlier iOS 26.4 version. Twenty turns is roughly the point where an assistant stops losing the thread halfway through a task, which is the specific failure that trained a generation of iPhone owners not to bother asking.

Coverage is broad. Siri runs across iPhone, iPad, Mac, Apple Watch, CarPlay, AirPods, Apple TV and Vision Pro. iPhone 15 Pro and newer, plus M1 or later iPads and Macs, get priority access. If you have a HomePod mini sitting on a shelf, the assistant it wakes is now the one Google helped build.

The hardware that arrived with it

Alongside the beta, Apple announced the Watch Series 12 at 399 dollars with a feature set it calls Audio Intelligence. That includes Live Rewind and Siri Recap, both built on ambient listening rather than explicit commands. An assistant that is always half listening is a different product from one that waits for a wake word, and the privacy documentation is worth reading before deciding how you feel about it.

Why the Google arrangement matters

Apple spent years making a virtue of doing AI on its own terms, on device where possible, with its own models. It then licensed the core of its flagship assistant from the company it competes with most directly in mobile. That is not quite a failure. Frontier model training is enormously expensive, and the gap between Apple's in-house work and the leading labs was visible to anyone using both. The same economics are reshaping who partners with whom across the industry, which is the backdrop to the race between the big labs toward public markets and the benchmarks the newest frontier models post.

It also leaves Apple shipping a product whose most important component it does not control. Terms change, pricing changes, and the model roadmap belongs to someone else. Industry efforts such as the Open Secure AI Alliance exist partly because that dependency now applies to almost everyone building consumer AI.

What to watch next

The open question is behavioural rather than technical. Siri works better than it did. The habit of not asking Siri anything is about a decade old, and habits like that do not reverse because the model underneath improved. Watch whether daily use of the standalone app holds up once the novelty wears off, because that is the number Apple will care about and the one least likely to turn up in a press release.

Some links in this article are affiliate links. We may earn a small commission at no extra cost to you.