The strangest thing about the Siri you’ll be talking to this fall is that, under the hood, you’re talking to Google.
At WWDC on June 8, Apple finally showed the rebuilt Siri it’s been promising since 2024. The headline feature isn’t the new design or the dedicated app — it’s that the cloud brain behind it is a custom, roughly 1.2-trillion-parameter Gemini model that Apple licenses from Google for an estimated $1 billion a year. Apple tunes it, wraps it in its own privacy plumbing, and slaps the Siri name on it. You will never see the word “Gemini” on your screen.
So now there are three serious AI assistants living on your iPhone, and one of them is quietly built on another. That changes the math on which apps are worth keeping. Let me walk through what actually shipped, how the three stack up by the job you’re trying to get done, and where the privacy story gets genuinely interesting.
What Apple actually changed
For years Siri was a command parser with a thin voice on top. It set timers, butchered your texts, and bailed to a web search the moment you asked anything real. The iOS 27 version is a different animal.
Siri is now a standalone app. It has a conversational interface with persistent chat history, so you can follow up without re-explaining yourself. It plugs into the Dynamic Island, reads on-screen context, and — this is the part that matters — it can reach across your apps. The demo Apple keeps showing: get a restaurant rec in a text, and Siri can pull the name, check your calendar for a free evening, and book the reservation in one go. You can upload files and photos to it now, too.
Three things power this. Simple stuff runs on-device with Apple’s own foundation models. Anything heavier routes to the cloud Gemini model through Apple’s Private Cloud Compute. And the whole thing has access to your “personal context” — emails, photos, files, messages — to actually personalize answers instead of pretending it doesn’t know who you are.
The catch on hardware: the full Gemini-powered experience needs an iPhone 15 Pro or newer with an A17 Pro chip and at least 8GB of RAM. Older supported phones get the redesign and minor improvements but not the smart cloud brain. iOS 27 ships publicly in September, with public betas running through the summer.
Three assistants, three different jobs
Here’s where people get the comparison wrong. They ask “which is best” like it’s one race. It isn’t. Each of these is now genuinely good at a different thing, and the smart move is matching the tool to the task instead of crowning a winner.
Siri owns your phone
Nothing else on this list can actually do things on your iPhone. ChatGPT can write you a brilliant email; it can’t open Mail, find the thread, and send it. Siri can. That’s the entire pitch, and it’s a real one.
On-screen awareness, cross-app actions, device control, “remind me about this when I get home” — this is Siri’s turf because Apple controls the operating system and the other two don’t. If your request involves your actual stuff (your calendar, your photos, your messages) and a follow-up action, Siri is now the obvious choice. It wasn’t a year ago.
What I’m skeptical about is reliability. Apple has a long history of demoing seamless multi-step Siri flows that fall apart the second your data is messy or your phrasing is off. The architecture is finally there. Whether the execution holds up across a million real inboxes is the open question, and we won’t really know until the thing is in people’s hands for a few months.
ChatGPT is still the one for hard thinking
When the task is the work itself — drafting something long, reasoning through a problem, writing and debugging code, building a custom workflow — ChatGPT Plus is still my default, and the benchmarks back that up. GPT-5.4 scores 71.7% on SWE-bench versus Gemini 3.1 Pro’s 63.8%. That gap shows up in practice on anything code-shaped or multi-step.
Custom GPTs, the agent features, the depth of its reasoning on a gnarly prompt — none of that is going anywhere just because Siri got smarter. ChatGPT Plus runs $20/month, and if you write, code, or think for a living, it’s doing a different job than your phone assistant. Siri booking a dinner reservation and GPT-5.4 refactoring your function are not competing.
Gemini is the research and Workspace play
Gemini’s edge is live search and sheer multimodal range. It’ll ingest up to an hour of video or eight hours of audio in a single input, which is absurd and occasionally exactly what you need. For current-events research, pulling from across the live web, or summarizing a long video, it’s the strongest of the three.
The other reason to keep it: Workspace. If you live in Gmail, Docs, and Sheets, Gemini is baked directly into them — summarize a thread without leaving Gmail, analyze a spreadsheet inside Sheets. Google AI Pro (the rebranded Gemini Advanced) is $19.99/month and bundles 2TB of Google One storage, which quietly makes it the best-value subscription of the three if you were paying for storage anyway.
There’s a delicious irony here. The same Gemini model powering Siri is the one Google sells you directly — except in Siri it’s white-labeled, tuned by Apple, and walled off from the live web in ways the standalone app isn’t. Same engine, very different car.
One more thing worth flagging on Gemini: the custom model Apple licensed is reportedly eight times larger than Apple’s own 150-billion-parameter cloud models, using a mixture-of-experts design tuned for summarizing, planning, and language understanding. So when Siri suddenly feels smart this fall, that’s the reason — it’s not Apple’s model that leveled up, it’s the one it rented.
The privacy story is the real plot twist
This is the question I kept getting from friends: if Siri runs on Google’s model, is Google now reading my email?
Apple’s answer is a flat no, and the architecture is built to make that more than a promise. Queries that go to the cloud run through Private Cloud Compute, which Apple says doesn’t store your data, doesn’t use it to train anything, and isn’t accessible to Apple employees. Crucially, Google is contractually barred from training on your queries, and the requests are anonymized and tokenized before they ever hit the model. The model runs on Apple-controlled infrastructure, not Google’s servers.
Compare that to using the Gemini or ChatGPT apps directly. There, by default, your conversations can be used to improve the models unless you dig into settings and turn it off. So the genuinely counterintuitive result is that Gemini-inside-Siri may be the most private way to use a Gemini-class model on your phone — more private than Google’s own app.
I’d add the usual caveat: this all rests on Apple’s claims and its contract with Google, not on something you can personally audit. Private Cloud Compute does publish its software images for security researchers to inspect, which is more than almost anyone else offers, and Apple’s commercial incentives are genuinely aligned with privacy in a way Google’s aren’t. That’s about as good as the trust situation gets in 2026. But it is still trust.
So should you delete the ChatGPT and Gemini apps?
Short answer: probably not, but it depends on what you actually do.
Delete both, lean on Siri — if you mostly use AI for quick questions, reminders, summarizing what’s on your screen, and small tasks involving your own apps and data. The new Siri genuinely covers this, it’s free, and it’s the most private option. The casual user who was paying $20/month out of FOMO can stop.
Keep ChatGPT Plus — if you write, code, or do anything that needs deep reasoning or a long output. Siri isn’t built for the heavy lift, and the GPT-5.4 quality gap is real on hard tasks. This is the one subscription I’d keep before any other.
Keep Gemini / Google AI Pro — if you live in Google Workspace or do a lot of web research and video work. The Workspace integration and the 2TB storage bundle make it pay for itself for the right person. If you’re not in that bucket, Siri’s Gemini brain already gives you most of the model quality for free.
Keep all three — if you’re a power user who context-switches constantly. They don’t overlap as much as the marketing implies. I keep all three and reach for each one without thinking about it, the same way I don’t agonize over which kitchen knife to grab.
The thing to stop doing is paying for an assistant you only use for tasks the free, built-in one now handles. A lot of people are in that spot and haven’t noticed yet.
What’s still missing
A few honest gaps before you get too excited.
The full experience is gated behind an iPhone 15 Pro or newer. If you’re on anything older, the September update is mostly a fresh coat of paint. That’s a hard pill given how Apple marketed “Apple Intelligence” as a platform-wide leap.
There’s also the dependency itself. Apple is now paying a rival a billion dollars a year to power its flagship assistant, which is a strange admission of where Apple’s own models landed. It works for now, but it’s a band-aid on a bigger problem, and the long-term plan to build its way out of that arrangement is anyone’s guess. The deal reportedly has an out clause if Apple’s in-house models catch up — which tells you Apple sees this as temporary too.
And the multi-step magic still has to survive contact with reality. The reservation demo is lovely. The version that runs against your chaotic calendar and your three half-finished email drafts is the one that counts, and we don’t have that data yet.
If you’re on a recent iPhone, install the public beta this summer and throw your actual daily requests at it — not the clean demo prompts, the messy real ones. That’s the only test that’ll tell you which of these apps earns its spot on your home screen come September.
Sources: MacRumors — WWDC 2026 expectations, TechCrunch — Gemini to power Apple’s AI, Apple Newsroom — Apple Intelligence 2026, 9to5Mac — Siri is powered by Gemini but isn’t Gemini, Demandsage — Gemini Advanced vs ChatGPT Plus.