new siri ai: not quite a loss to google

apple promised a "more personal siri" back at wwdc 2024, missed it, delayed it through 2025, missed it again, and spent most of last year quietly walking back the timeline. now, at wwdc 26, it's finally here – siri ai. and it's genuinely good. but there's a moment. here's the breakdown

core capabilities

siri ai is built around four things it can now do that old siri couldn't:

• personal context – searches across messages, mail, photos, and third-party apps (via spotlight integration) to find things like a friend's restaurant rec or an old hotel confirmation
• onscreen awareness – answers questions about whatever is currently on your screen
• web knowledge – pulls up-to-date info from the web; supports follow-up questions
• systemwide app actions – drafts emails, edits/shares photos, etc., across apps

new surfaces

beyond what it can do, apple also rebuilt where you reach it:

• dedicated siri app with icloud-synced conversation history across devices
• spotlight integration on ipad and mac
• dynamic island swipe-down on iphone
• 3d visualization on vision pro you can place in space
• available on apple watch, carplay, airpods

visual intelligence and writing tools

visual intelligence expanded to ipad, mac, and vision pro. new siri mode in camera on iphone – split bills via apple cash, get nutrition info, etc.

writing tools – generate drafts, refine text, match your tone per recipient in mail/messages, systemwide proofreading (works in most third-party apps too).

under the hood

1. on-device model improvements: more expressive/customizable voices, much better dictation with auto punctuation and formatting.
2. privacy architecture: runs on-device or via private cloud compute (data not stored or accessible to apple); orchestrator and spotlight index stay on-device.

but here's the part apple downplayed in the keynote: the heavy lifting isn't actually apple's own model.

in january 2026, apple and google announced a multi-year partnership where the next generation of apple foundation models is based on google's gemini. the custom gemini model built for siri is reportedly around 1.2 trillion parameters using a mixture-of-experts architecture – roughly eight times larger than apple's previous cloud models. apple's own framing: "after careful evaluation, we determined that google's technology provides the most capable foundation for apple foundation models."

so the split is roughly:

• on-device tasks – apple's smaller foundation models on the neural engine
• heavier reasoning, summarization, planning – the custom gemini model, executed via private cloud compute

availability

• developer testing today on ios 27, ipados 27, macos 27, visionos 27 (watchos 27 later)
• public beta later this year, english first
• requires iphone 16+, iphone 15 pro/pro max, m1+ ipads/macs, vision pro, apple watch series 10/ultra 2/se 3
• eu – mac/watch/vision pro yes; not initially on iphone/ipad
• china – not available yet (regulatory)

vs gemini on android

where gemini is ahead:

• shipping now, not "later this year." gemini already replaced google assistant on most android phones, with gemini intelligence rolling out across pixel 10 and galaxy s26
• more mature agentic actions – multi-step tasks across apps (reserve table, set reminder, send follow-up text) are live today
• broader reach – android, wear os, android auto, google tv, even budget hardware via gemini go
• available in the eu and most regions on day one

where siri ai is ahead:

• privacy architecture – private cloud compute with external verifiability is genuinely stronger than gemini's mostly cloud-based processing
• deeper os integration on apple devices – dynamic island, spotlight, system context menus, vision pro spatial ui, cross-device sync via icloud
• writing tools that mimic your tone per recipient
• on-device dictation and voice customization

conclusion

the gemini partnership reshapes how to read this announcement. siri ai's intelligence ceiling is essentially gemini's ceiling – which is high, but means apple isn't bringing new model capability to the market, it's bringing new packaging. that packaging is real and valuable: private cloud compute is a meaningfully better privacy posture than what google offers android users, and apple's os-level integration is tighter than anything gemini can do on android without owning the stack.

but the strategic picture is less flattering. analysts have read the deal as apple conceding it couldn't build a competitive frontier model in-house after years of trying – ming-chi kuo called it "a way to ease short-term pressure rather than a long-term strategic shift." apple delayed its personalized siri overhaul through most of 2025, and the gemini deal is what unblocked the ship date.

so: is siri ai better than gemini on android? right now, no – gemini ships today, has more agentic surface area, and runs on more hardware. once siri ai reaches public beta, the more honest framing is that they're the same brain in different bodies, and the choice comes down to which ecosystem and which privacy model you prefer. apple's track record of announcing siri features that arrive late or under-deliver (the 2024 "more personal siri" slipped badly) is reason to wait for the beta before declaring a winner.