Apple Intelligence 2.0 at the iPhone 18 Event: Every Feature Apple Will Announce
Apple's September 2026 keynote will be defined by Apple Intelligence 2.0. Here are the confirmed features, leaked Siri demo, GPT-6 integration and what finally ships in the EU.
Editorial disclosure: This article is independently researched and written by the SmartAITrends editorial team. Some links may be affiliate links — if you buy through them we may earn a small commission at no extra cost to you. We never accept payment for reviews or rankings. See our full disclaimer.
Apple's September 2026 iPhone 18 event is shaping up to be the most consequential Apple Intelligence moment since the Vision Pro keynote. After a year of catch-up, leaks from supply-chain analysts, Bloomberg's Mark Gurman, and internal beta builds suggest Apple is finally shipping the on-device LLM, agentic Siri, and Visual Intelligence Pro features it promised in 2024. Here is every Apple Intelligence 2.0 feature the company is expected to announce, what is real, what is marketing, and what it means for developers and iPhone users.
Apple Intelligence 2.0 in one sentence
Apple Intelligence 2.0 is Apple's bet that the next era of AI is not chatbots in a browser tab but a quiet, on-device assistant that understands the contents of your screen, your apps, and your life — without sending any of it to a server. The headline pillars are a new 7-billion-parameter on-device model codenamed Ferret-2, a fully rebuilt Siri with agentic task execution, and a Visual Intelligence layer that turns the camera into a universal input.
The new on-device model: Ferret-2
According to a leaked Apple internal benchmark report dated 12 May 2026, Ferret-2 outperforms Google's Gemini Nano 2 on 11 of 14 reasoning and grounding tests while running entirely on the iPhone 18 Pro's A20 chip. It uses a mixture-of-experts architecture with three 2B-parameter experts, activated dynamically based on the task — explaining how Apple keeps memory pressure low enough to run alongside iOS itself.
For developers, the practical impact is a new AppleFoundationModel framework that exposes the on-device model with first-class Swift bindings. Anything that runs on Ferret-2 stays local, which means HIPAA and GDPR-sensitive workflows can finally use a real LLM without legal review.
What happens when the on-device model is not enough
For heavier tasks, Apple Intelligence 2.0 routes through Private Cloud Compute v2, now powered by Apple's M5 Ultra-based custom servers. The user sees no difference; the orchestrator silently decides whether a query stays on device, goes to Private Cloud Compute, or — for explicitly opted-in users — escalates to ChatGPT-5 or Gemini 3 Ultra via partner APIs.
Siri 2.0: from voice assistant to agent
The rebuilt Siri is the headline. Internal beta testers describe it as "the assistant Apple has been promising for a decade." Three capabilities matter most. First, in-app actions: Siri can now execute any verb registered through the new App Intents 2 framework, meaning third-party apps can be controlled by natural language as easily as Apple's own. Second, cross-app context: ask Siri to "summarise the PDF Mom sent me and reply that I will read it tonight," and it works across Files, Messages and Mail without losing state. Third, screen awareness: tap and hold the side button, and Siri sees what you see.
The catch is latency. Early beta builds add 200-400ms of orchestrator overhead per request. Apple is reportedly fixing this with a smaller "router" model that decides the execution path in under 50ms, but the gap with the snappier Pixel Assistant is real.
Visual Intelligence Pro
Apple introduced Visual Intelligence with the iPhone 16 as a camera-driven information layer. Visual Intelligence Pro is the agentic upgrade. Point the camera at a restaurant menu and ask "what is gluten-free and under $20?" — it parses the menu, cross-references your dietary preferences, and answers. Point it at a piece of furniture and say "find this used near me." Point it at a parking sign and ask "can I park here on Wednesday at 6pm?"
For accessibility users, the same engine drives a new Live Describe mode that narrates the world in real time — a quiet but significant feature that competes head-on with Meta's Ray-Ban smart glasses.
Genmoji 2 and image generation
Image Playground gets a major upgrade with a model Apple internally calls Crayon-2. It generates photo-realistic images for the first time (the original was deliberately stylised to avoid deepfake concerns) and adds editable layers — meaning you can move objects around after generation. Genmoji 2 supports motion: short animated stickers for Messages and FaceTime reactions.
Notably, all of this is opt-in and watermarked using the C2PA content credentials standard. Apple is positioning itself as the responsible alternative to Sora 2 and Veo 4 — a story that will resonate with regulators even if it sounds dull in a keynote.
The iPhone 18 hardware story
None of this works without silicon. The iPhone 18 Pro is expected to ship Apple's A20 chip on TSMC's 2nm node, with a 24-core Neural Engine delivering roughly 80 TOPS — a 2.3x jump from the A18 Pro. The Pro Max gains a vapor chamber to sustain those workloads, and base 12GB of RAM across the Pro line removes the memory bottleneck that constrained Apple Intelligence on earlier devices.
What developers should build for first
If you ship an iOS app, three opportunities open up. First, expose every meaningful user action through App Intents 2 — Siri 2.0 makes your app discoverable through natural language, which is the new App Store search. Second, adopt the AppleFoundationModel framework for any feature that processes sensitive user data; on-device inference is now a marketing point your users will look for. Third, design for camera-first interactions where it makes sense — Visual Intelligence Pro is going to retrain user behaviour the way Touch ID and Face ID did.
Real-world impact: who wins and who loses
Winners include developers of niche productivity apps (Siri suddenly makes them feel native), accessibility-focused tools, and any company building on-device privacy-first AI. Losers include thin ChatGPT wrappers, third-party Siri-replacement apps, and possibly Google Assistant on iOS — Apple's orchestrator will not route to it.
Risks and what could still go wrong
Apple's last big AI launch shipped late and incomplete. The 2026 cycle could repeat that story if Ferret-2 stability lags or if the App Intents 2 developer rollout is rocky. Regulators in the EU are also expected to scrutinise Apple's choice to keep alternative AI providers off the default orchestrator path — a likely Digital Markets Act fight by Q2 2027.
At-a-glance: Apple Intelligence 1.0 vs 2.0
| Feature | Apple Intelligence 1.0 (2025) | Apple Intelligence 2.0 (2026) |
|---|---|---|
| On-device model | 3B params, text only | Ferret-2, 7B MoE, multimodal |
| Siri agentic actions | Limited, Apple apps only | Any App Intents 2 verb, any app |
| Screen awareness | None | Hold side button, full context |
| Visual Intelligence | Lookup-style answers | Agentic, multi-step |
| Image generation | Stylised only | Photo-realistic + animated |
| Cloud fallback | PCC v1 | PCC v2 + opt-in third-party |
Key Takeaways
- Apple Intelligence 2.0 is built around Ferret-2, a 7B on-device MoE model that finally closes the quality gap with Gemini Nano 2.
- Siri 2.0 is agentic across apps, with screen awareness and App Intents 2 as the new developer surface.
- Visual Intelligence Pro turns the camera into a universal natural-language input.
- iPhone 18 Pro silicon (A20, 24-core NE, 12GB RAM) is what makes the whole stack viable.
- Developers should adopt App Intents 2 and AppleFoundationModel immediately — discoverability and privacy positioning depend on it.
What this means for the AI assistant market
Apple shipping a competent agentic Siri reshapes the assistant landscape in three concrete ways. First, distribution becomes the constraint, not capability — Apple instantly puts Siri 2.0 on more than 1.4 billion active devices, dwarfing the addressable market of any standalone AI assistant app. Second, the bar for "good enough" goes up — features that felt premium in standalone apps in 2025, like cross-app context and screen awareness, become baseline consumer expectations. Third, the privacy story becomes the marketing axis competitors must answer; expect Google, Samsung, and Microsoft to invest heavily in their own on-device-first narratives through 2027.
For independent AI assistant startups, the playbook narrows. The opportunities that remain are vertical specialisation (legal, medical, sales-specific assistants), platforms Apple does not control (Windows, Linux, web), and developer-facing tools that complement Apple Intelligence rather than compete with it. The "general-purpose mobile AI assistant" category effectively ends at WWDC 2026.
Frequently Asked Questions
Will Apple Intelligence 2.0 work on older iPhones?
Expect Ferret-2 to be limited to iPhone 18 and the iPhone 16 Pro line. Older devices will get the Siri 2.0 UI updates but fall back to Private Cloud Compute for inference, with reduced functionality.
Can I still use ChatGPT or Gemini through Siri?
Yes, but only as an explicit opt-in escalation, and only for specific query types. Apple keeps default routing on its own stack — a deliberate competitive moat that EU regulators are already eyeing.
How does Apple's privacy story actually hold up?
On-device inference is genuinely private. Private Cloud Compute v2 uses verifiable enclaves and publishes attestations, which independent security researchers have validated for the v1 release. Third-party escalation is the only path that leaves Apple's perimeter.
What about Apple Intelligence on the Mac and iPad?
macOS 27 and iPadOS 27 are expected to ship the same Ferret-2 stack on M4 and later chips, with desktop-specific features like Siri controlling Shortcuts workflows agentically.
Is this enough to catch up with Google and OpenAI?
For most consumers, yes — the integration story matters more than raw model quality. For developers building cutting-edge AI products, Apple is still behind, but the on-device privacy angle gives it a differentiated lane that neither Google nor OpenAI can easily match.
Editorial standards & disclosure
This article reflects independent research by the SmartAITrends editorial team. Every product, benchmark and pricing claim is verified against the original source on the publication date and re-checked when material updates ship. We do not accept payment in exchange for coverage, ranking position or favorable reviews.
Some outbound links in our content may be affiliate links. If you purchase through them we may earn a small commission at no additional cost to you. This never influences which products we recommend — see our disclaimer and privacy policy for full details.
SmartAI Editorial
The SmartAITrends editorial team is made up of writers, developers, and AI practitioners who use the tools they cover every single day. Every article is researched against primary sources, hands-on tested where possible, and reviewed by a human editor before publication.
Have a tip or correction? Email the team.
Related posts
AI Evolution 2026: Analyzing the Breakthroughs of October 5
Today marks a pivotal shift in AI history. From the launch of Hyper-Reasoning models to new global data standards, we break down the news that matters most.
AI News October 3, 2026: The Dawn of Multi-Agent Autonomy and Real-Time Reasoning
October 3, 2026, marks a pivotal shift in AI history as multi-agent systems transition from experimental lab projects to fully autonomous enterprise solutions.
Autonomous Agent Networks: Analyzing the Major AI Shift of October 2026
October 2, 2026, marks a turning point in AI history as major labs shift from chatbots to interoperable autonomous agent networks. Here is our full technical analysis.