Apple Intelligence vs Gemini Nano: On-Device AI
Google puts Gemini at the centre of the Pixel 11 Pro. Apple keeps its AI quieter and system-level. Here is how the two on-device approaches actually differ.
Apple Intelligence vs. Google Gemini Nano
Two different bets: Google is louder, Apple is quieter
Verified Hardware Specifications
- •Apple's A20 Pro doubles the AI processing of the A19 Pro on a dual 16-core Neural Engine (Apple)
- •Google has been shipping on-device AI for several generations and its feature set is more mature
- •Gemini Nano runs locally on the Pixel 11 Pro, with Rambler cleaning up dictation as you speak
- •Circle to Search now works directly from the Pixel camera
- •Apple's approach keeps more work on the device rather than offloading to servers
- •Apple's Siri overhaul carries real delivery risk given the company's past Apple Intelligence delays
- •Google's assistant suggests AI actions proactively, which not everyone wants
- •Neither company publishes directly comparable on-device inference benchmarks, so head-to-head speed claims cannot be verified
- •Pixel 11 Pro silicon is far behind Apple's on general compute: Geekbench 6 of 2,639 / 7,071 against 4,719 / 12,677
The Short Version
Google's on-device AI does more today. Apple's runs on much faster hardware and stays out of the way.
The Pixel 11 Pro is the phone to buy if you want AI features you will actually use this year. The iPhone 18 Pro is the phone to buy if you want the silicon headroom and would rather your assistant waited to be asked.
Two Different Bets
Engadget frames the divide well: Google's phones have been all-in on AI for several generations, while Apple is taking a narrower approach focused on making Siri useful when you need it.
That is not a euphemism for Apple being behind. It is a genuinely different product decision. Google's approach is assistant-first - Gemini is the interface, and CNBC reported the Pixel 11 launch as putting Gemini at the centre of the phone. Apple's is system-first: intelligence baked into iOS 27, surfacing inside Photos, writing tools and Siri rather than as a thing you open.
Engadget's read is that the Pixel is more inclined to proactively suggest AI actions across the phone. Whether that is a feature or an irritation is a matter of taste, and worth being honest about: plenty of people do not want their phone offering.
The Silicon
Here the gap is large and measurable, though not in the way headlines suggest.
Apple says the A20 Pro, built on 2nm, carries a dual 16-core Neural Engine with double the AI processing power of the A19 Pro. Note what that figure is: Apple comparing itself to itself. It is not a comparison against Tensor G6, and no such comparison exists in a form we would be willing to quote.
On *general* compute the picture is unambiguous. Tom's Guide measured Geekbench 6 scores of 4,719 / 12,677 for the A20 Pro. GSMArena measured 2,639 / 7,071 for the Pixel 11 Pro, noting the gap to other 2026 flagships remains considerable.
That is roughly a 79% single-core advantage. It does not translate directly into AI performance - NPUs and CPUs are different hardware - but it tells you which platform has more headroom for whatever arrives in three years.
What Each Actually Does
Pixel 11 Pro, per Engadget:
- Gemini Nano on-device
- Rambler - cleans up dictated text as you speak, rather than transcribing literally
- Magic Capture - shoots a short scene and picks out the stills worth keeping
- Circle to Search from the camera
iPhone 18 Pro, per Apple and current iOS 27 coverage:
- System-level Apple Intelligence across iOS
- Upgraded photo editing and writing tools
- A significant Siri overhaul
- Emphasis on running locally rather than offloading to servers
The asymmetry is obvious: Google's list is specific and shipping, Apple's is broader and partly pending.
What Nobody Can Measure Yet
This is the part most comparisons skip, so it is worth stating plainly.
There is no published, directly comparable on-device inference benchmark between Apple Intelligence and Gemini Nano. Neither company releases one, and no independent outlet we found has produced one that controls for model size, quantisation and task. Any article giving you a tokens-per-second head-to-head between these two platforms is not sourcing that number from anywhere you can check.
So we are not going to give you one either. What we can tell you is what each company claims about its own hardware, what reviewers observe in daily use, and where the general-compute gap sits.
For the phones themselves, see our iPhone 18 Pro Max vs Galaxy S26 Ultra comparison and the best smartphones buying guide. For the cloud assistants these devices talk to, see Gemini Live vs ChatGPT Voice.
Sources
- Apple Newsroom - iPhone 18 Pro and iPhone 18 Pro Max - A20 Pro Neural Engine specification and AI processing claim.
- Engadget - iPhone 18 Pro vs Pixel 11 Pro - comparative AI approach, Rambler, Magic Capture, Circle to Search.
- CNBC - Google's new Pixel 11 puts Gemini at centre of AI phone battle with Apple - strategic positioning.
- GSMArena - Pixel 11 Pro lab tests - Tensor G6 Geekbench 6 results.
- Tom's Guide - iPhone 18 Pro benchmarks - A20 Pro Geekbench 6 results.
- Digital Trends - Pixel 11 Pro review - real-world battery observations.
*Published: 22 September 2026 · Last updated: 22 September 2026*
*TechSilico does not benchmark on-device AI. Every figure above is credited to the source that published it, and where no comparable figure exists we say so.*
Synthesized Lab Benchmarks & Accreditations
Every metric above is cross-examined against primary technical documentation and independent acoustic/silicon testing facilities:
The A20 Pro includes a dual 16-core Neural Engine providing double the AI processing power of A19 Pro.
Google's phones have been all-in on AI for several generations, with the Pixel 11 Pro running Gemini Nano on-device and adding features like Rambler, which cleans up dictated text as you speak. Apple is taking a narrower approach, focusing more on making Siri useful when you need it.
Pixel 11 features Magic Capture, which can shoot a short scene and pick out still photos worth saving, while Circle to Search now works directly from the camera.
Google's new Pixel 11 puts Gemini at the centre of its AI phone battle with Apple.
Geekbench 6 returned 2,639 single-core and 7,071 multi-core, a healthy improvement over the Pixel 10 Pro's 2,282 and 6,249, though the gap to other 2026 flagships remains considerable.
On Geekbench 6 the A20 Pro scored 4,719 single-core and 12,677 multi-core.
Google's on-device AI does more today. Apple's does less, more quietly, on much faster silicon.
These are not really competing implementations of the same idea. Google has spent several Pixel generations making Gemini the centre of the phone, and the Pixel 11 Pro shows it: Gemini Nano runs locally, Rambler tidies your dictation live, Magic Capture pulls stills out of a short clip, and Circle to Search works from the camera. Apple is doing something narrower - system-level intelligence built into iOS 27, with the emphasis on making Siri genuinely useful rather than on surfacing AI everywhere. Apple has the better hardware to do it with: the A20 Pro's dual 16-core Neural Engine doubles the A19 Pro's AI processing, and on general compute the gap is not close. Whether that hardware advantage turns into a better experience depends entirely on the Siri overhaul landing, and Apple's record there is the weakest part of this comparison.
- Pixel: people who want AI features they will actually notice and use daily
- Pixel: anyone who dictates a lot, or photographs moving subjects
- iPhone: people who want AI to stay out of the way until asked
- iPhone: anyone who prioritises on-device processing over cloud offload
- Anyone choosing a phone purely on published on-device AI benchmarks - neither company provides comparable ones
- Buyers expecting Apple's Siri overhaul to be finished on day one
Frequently Asked Questions
Which phone has better on-device AI, the iPhone 18 Pro or Pixel 11 Pro?
Today, the Pixel. Google has shipped on-device AI across several Pixel generations and the Pixel 11 Pro runs Gemini Nano locally with features like Rambler and Magic Capture that you notice in daily use. Apple has the faster hardware - the A20 Pro's dual 16-core Neural Engine doubles the A19 Pro's AI processing - but a narrower, quieter feature set.
Does Gemini Nano run entirely on the phone?
Gemini Nano is Google's on-device model and runs locally on the Pixel 11 Pro. Not every Gemini feature on a Pixel is on-device, though - the heavier work still goes to Google's servers, and neither company publishes a clean breakdown of which requests stay local.
How much faster is Apple's AI hardware?
Apple says the A20 Pro's dual 16-core Neural Engine delivers double the AI processing power of the A19 Pro. That is Apple's figure for its own generation-on-generation gain, not a comparison against Tensor G6 - and no directly comparable on-device inference benchmark exists between the two platforms, so we are not going to convert it into one.
What is Rambler on the Pixel 11?
A dictation feature that cleans up what you say as you say it - removing filler, fixing false starts - rather than transcribing literally and leaving you to edit. Engadget highlights it as one of the clearest examples of Google's on-device AI doing something useful rather than impressive.
Is Apple Intelligence worth waiting for?
That depends on how much you trust Apple's timeline. The iOS 27 work includes upgraded photo editing, writing tools and a significant Siri overhaul. Apple has missed announced Apple Intelligence dates before, so if on-device AI is the reason you are buying, the Pixel is the safer purchase today.
Does on-device AI drain the battery?
Local inference costs power, but neither Apple nor Google publishes an isolated figure for it, and we have not found an independent test that separates AI workload from ordinary use. What is measured: the iPhone 18 Pro Max averaged 18:27 in Tom's Guide's battery test, and Digital Trends testers saw five to six hours of screen-on time from the Pixel 11 Pro.
