Apple Intelligence 2.0 Review (2026): What's New, What Works

Apple Intelligence launched in late 2024 as a polite, half-finished promise. Two years later, Apple Intelligence 2.0 is finally the thing the keynote sold us — and yet the most honest review you'll read this year still has to say "it depends." Here's what's new, what actually works in 2026, and where Apple is still losing to Gemini and Galaxy AI.

What's New in Apple Intelligence 2.0

The original Apple Intelligence rollout was a slow drip — Writing Tools first, then Image Playground, then a Siri that could finally see your screen, all spread across point releases of iOS 18. The 2.0 release that shipped with iOS 19 in spring 2026 collapses all of that into a single coherent product, and adds three things the original was missing: a real on-device large language model, true contextual memory across apps, and a developer framework that lets third-party apps plug into the same intelligence layer.

The headline upgrades:

• Apple Foundation Model (on-device) — a 3B-parameter LLM that runs entirely on the Neural Engine, fine-tuned for everyday personal tasks (summarization, rewriting, classification, simple Q&A). No cloud round-trip, no data leaving the device.
• Siri 2.0 — full conversational memory across a session, screen-aware actions, and the ability to chain commands across apps without the awkward "wait for the prompt" pause.
• Writing Tools v2 — tone presets, style memory, and a "rewrite in my voice" mode that learns from your last 90 days of Messages and Mail (on-device only).
• Image Playground 2 — photorealistic mode (finally), better text rendering, and a "make it match my notes" feature that pulls aesthetic cues from your own photo library.
• Genmoji 2 — multi-character scenes, animation, and Sticker Pack export.
• Private Cloud Compute 2.0 — when a task escapes the on-device model, it now routes to Apple-controlled servers with verifiable transcript privacy, and you can audit which queries went where.
• App Intents 2 — third-party apps can register actions, data types, and entities that Siri and the system intelligence can use directly.

None of this is a leap into the future. It's Apple shipping what they promised eighteen months ago, polished into something that mostly just works. That, in 2026, is rarer than you'd expect.

The 6 Killer Features (Tested for Two Weeks)

I've been running Apple Intelligence 2.0 daily on an iPhone 16 Pro Max and an M4 MacBook Pro for two weeks. Here's the honest breakdown of what's actually useful versus what still feels like a demo.

1. Siri 2.0 — Finally a Real Assistant

This is the upgrade most people will feel first. Siri now remembers the conversation — you can say "open Maps and find a coffee shop nearby," then say "the second one, share it with Sarah," and it works. No re-prompting, no "I didn't catch that." Screen awareness means you can highlight an address on a webpage and say "add this to my contacts" without copy-pasting.

It's still not as smart as ChatGPT for open-ended questions — Apple routes those to the optional ChatGPT integration — but for device-level tasks, Siri finally feels like an assistant instead of a voice-activated remote.

2. Writing Tools v2 — Grammarly's Real Competition

The new Writing Tools panel lives one tap away in every text field across iOS and macOS. You can rewrite a sentence in five tones (Professional, Friendly, Concise, Confident, Casual), summarize a long email into bullet points, or transform a rough draft into a proper outline. The killer addition is "Rewrite in my voice" — it analyzes your historical writing on-device and matches your tone, sentence length, and vocabulary.

I tested it against Grammarly Premium and Notion AI on the same 800-word draft. Apple's output felt the most like me, Grammarly's felt the most grammatically polished, and Notion's was the most creative but the most generic. For daily writing inside Mail, Messages, and Notes, Writing Tools is now genuinely better than the alternatives.

3. Image Playground 2 — No Longer a Toy

The original Image Playground was so cartoony it felt like an Apple Watch app. Version 2 adds a real photorealistic mode and dramatically better text rendering inside images, which means it can finally produce thumbnails, mood boards, and social posts that don't scream "made with the default app." It's not Midjourney — the prompts still get filtered aggressively and the output ceiling is lower — but for personal use, it's the most frictionless image generator on any phone, period. Want a deeper read on where AI image generation is going? See our comparison of Gemini Deep Think and the multi-hypothesis reasoning that's pushing Google's visual models ahead.

4. Genmoji 2 — Genuinely Fun, Mildly Useful

Custom emoji generation got the upgrade nobody asked for but everyone enjoys. You can now build multi-character scenes ("me and my dog at the beach"), export them as Sticker Packs, and even create short animated versions. It's a feature that lives or dies on whether your friend group adopts it — but in the two iMessage threads where I've used it, response rates went up. Marketing-wise, that's not nothing.

5. On-Device LLM — The Quiet Game-Changer

The Apple Foundation Model is the feature Apple barely markets, and it's the one developers care about most. A 3B-parameter model running on the Neural Engine means every app on iOS 19 can call a real LLM without an API key, an internet connection, or a privacy review. Summarization, classification, rewriting, simple Q&A — all of it on-device, all of it free, all of it with the same privacy guarantees as the rest of iOS.

For third-party developers, this is the most important shift since the App Store. The next generation of journaling apps, email clients, and note-takers will be built on top of this model. Compared to what OpenAI is charging per million tokens — see our breakdown of GPT-5.5's pricing model — Apple effectively just made personal AI free for anyone shipping inside the App Store.

6. Privacy-First Architecture — The Real Moat

Every other AI assistant routes your queries to a cloud somewhere. Apple Intelligence runs roughly 80% of requests on-device, and when it does need the cloud, it uses Private Cloud Compute 2.0: ephemeral, verifiable, with no data retention and a public transparency log you can audit. Independent researchers have already published the auditing tools.

For users in regulated industries — healthcare, law, finance — this is the only consumer AI assistant that survives a serious compliance review. That's a real competitive moat, and it's one Google and Samsung can't easily copy without rebuilding their entire data infrastructure.

Apple Intelligence vs Gemini vs Galaxy AI

Two years in, the three big consumer AI stacks have settled into distinct personalities. Here's the side-by-side that matters in 2026:

The pattern is consistent: Gemini wins on raw intelligence ceiling, Galaxy AI wins on the niche super-features, Apple wins on integration and privacy. If you're already inside the Apple ecosystem, 2.0 is the upgrade that finally justifies the marketing. If you bounce between Android and iOS, Gemini is still the more powerful brain. The deeper story on Google's side — including how Gemini is now baked into Google Maps as a built-in assistant — shows how aggressively Google is closing the integration gap Apple used to own.

What Still Doesn't Work in Apple Intelligence 2.0

This is the section other reviews skip, so let's be direct. After two weeks of daily use, here's what's still rough:

• Siri still hands off too often. Anything beyond device control gets routed to ChatGPT — which works, but breaks the illusion of a single assistant. Gemini handles the same query natively.
• Image Playground prompts get over-filtered. Want to generate a moody noir scene? You'll fight the filter for ten minutes. Midjourney does it on the first try.
• Writing Tools "Rewrite in my voice" works in English only. If you write in Spanish, French, German, or Arabic, you get the standard model. Apple says multilingual voice matching is coming "later in 2026."
• Genmoji animations don't render on non-Apple devices. Send one to an Android friend and they get a static fallback. This is a deliberate design choice and it's frustrating.
• Older devices are locked out. The full Apple Intelligence 2.0 experience requires iPhone 16 Pro or newer, iPad with M2+, or Mac with M2+. Anything older gets a stripped-down version with no on-device LLM.
• No real "agent" mode. Siri still can't reliably chain multi-app workflows the way that, say, an Anthropic computer-use agent or a Galaxy AI routine can. Apple is clearly working on this — App Intents 2 is the foundation — but the consumer experience isn't there yet.

None of these are dealbreakers. All of them are reminders that Apple ships slowly and conservatively. If you're hoping for AGI on your phone, you'll still be disappointed. If you want a fast, private, well-integrated assistant that doesn't surprise you, 2.0 delivers.

Devices Supported

Here's the cleanest possible summary of the device matrix as of iOS 19.2 / macOS 16.2:

• Full Apple Intelligence 2.0: iPhone 16 Pro, 16 Pro Max, 17 (all variants), 17 Pro, 17 Pro Max. iPad Pro M2 and newer. iPad Air M2 and newer. All M2/M3/M4 Macs.
• Partial (no on-device LLM, no Image Playground photorealistic, limited Siri 2.0): iPhone 15 Pro, 15 Pro Max, 16, 16 Plus. iPad Pro M1. M1 Macs.
• No Apple Intelligence: iPhone 15 standard, all iPhone 14 and earlier, all Intel Macs, iPads with A-series chips below M1.

The cutoff is firm and Apple has not signalled any plans to back-port the full feature set. If you're on an older device, the upgrade pressure in late 2026 will be real.

Privacy & Data Handling — The Detail That Matters

Apple's privacy story is the most-marketed and least-understood part of this whole launch. Here's the actual architecture, in plain English:

1. On-device first. Roughly 80% of Apple Intelligence requests run entirely on your device. Nothing leaves. Apple cannot see the query, the result, or the context.
2. Private Cloud Compute 2.0 for harder requests. When the on-device model can't handle a task, the request is routed to Apple-controlled servers running custom Apple silicon. The servers do not retain data, do not log queries, and run software that is publicly auditable.
3. Transparency log. Every PCC server build is published with a cryptographic hash. Independent researchers verify that the software running matches the published code.
4. Optional ChatGPT routing. If you accept the integration, certain requests can be forwarded to OpenAI. You're prompted each time, and OpenAI's standard privacy terms apply.

This is the best consumer-AI privacy architecture currently shipping. It's not magic — if you opt into ChatGPT routing, you're back inside OpenAI's terms — but the default posture is the strongest in the market by a wide margin.

Who Should Upgrade — And Who Shouldn't

The honest verdict, split by user type:

• If you have an iPhone 15 Pro or older: the partial experience is fine but not transformative. Wait for the iPhone 17 cycle if you're due for a phone anyway.
• If you have an iPhone 16 Pro / 17 series: install iOS 19 today. The full Apple Intelligence 2.0 stack is the most polished consumer AI experience on any phone right now.
• If you're an Android user thinking of switching: don't switch for AI alone. Gemini is still the smarter raw model. Switch for ecosystem, privacy posture, or hardware — and treat Apple Intelligence as the bonus.
• If you're a developer: the App Intents 2 framework plus the free on-device LLM is the most important platform shift Apple has shipped since SwiftUI. Start building.
• If you're an enterprise IT lead: Apple Intelligence 2.0 is the first consumer AI stack you can actually deploy without rewriting your compliance docs. That alone is worth a procurement conversation.

The Final Take

Apple Intelligence 2.0 isn't trying to be the smartest AI. It's trying to be the most useful AI inside the world's most popular consumer ecosystem — and on that narrower goal, it succeeds. The on-device model removes the privacy excuse for staying on the cloud. Siri 2.0 finally stops being a punchline. Writing Tools beats third-party alternatives for daily use. And the developer story sets up the next two years of iOS app innovation.

The shortcomings are real — Siri still handing off to ChatGPT, Image Playground over-filtering, the device cutoff — but they're the kind of "Apple ships slowly and conservatively" shortcomings, not the kind of "this product is broken" shortcomings. Eighteen months from launch, Apple Intelligence is finally the thing the original keynote promised. That's worth saying out loud.

FAQ

Is Apple Intelligence 2.0 free?

Yes. All features are free for users on supported devices. The optional ChatGPT integration is also free for standard usage; ChatGPT Plus features require a separate OpenAI subscription.

Which iPhones support Apple Intelligence 2.0?

iPhone 16 Pro, 16 Pro Max, and all iPhone 17 models get the full experience. iPhone 15 Pro, 15 Pro Max, 16, and 16 Plus get a partial version without the on-device LLM. iPhone 15 standard and earlier are not supported.

Does Apple Intelligence work without an internet connection?

Most features do. The on-device Apple Foundation Model runs offline for summarization, rewriting, classification, and short Q&A. Heavier tasks route to Private Cloud Compute, which requires connectivity.

Is Apple Intelligence better than ChatGPT?

For device-level tasks (controlling apps, writing emails, summarizing notes) yes. For open-ended reasoning and long-form generation, ChatGPT (and the optional integration inside Apple Intelligence) is still stronger.

Can I disable Apple Intelligence?

Yes. Settings → Apple Intelligence & Siri → toggle off. Individual features (Writing Tools, Image Playground, Genmoji) can also be disabled separately.

Does Apple use my data to train its models?

No. Apple explicitly states that Apple Intelligence does not use personal data, queries, or content to train its foundation models. Private Cloud Compute does not retain request data.