Apple’s iPhone 18 Pro has demonstrated a major improvement in on-device AI performance, running a 27-billion-parameter model at roughly twice the speed of the iPhone 17 Pro.
The improvement comes from the A20 Pro, which introduces a dual-16-core Neural Engine in Apple Silicon, alongside faster memory in the iPhone 18 Pro and iPhone 18 Pro Max.
27B AI Model Runs Locally on iPhone
Adrien Grondin demonstrated Bonsai 27B running directly on an iPhone 18 Pro, showing significantly faster token generation than on the previous-generation device.
Both the iPhone 18 Pro and iPhone 18 Pro Max come with 12GB of 96-bit LPDDR5X memory, which is considerably faster than the memory configuration used in the iPhone 17 Pro series.
The iPhone 18 Pro also offers 115.2GB/s of unified memory bandwidth. Combined with the A20 Pro’s dual-16-core Neural Engine, this gives the phone considerably more resources for running AI models locally without an internet connection.
The Neural Engine’s peak throughput can also exceed the A20 Pro’s 7-core GPU in workloads specifically designed to run on the NPU.
The iPhone 18 Pro is much better than what I expected for on-device AI
It can run a 27B model at double the speed of the 17 Pro
The new A20 Pro chip is a beast pic.twitter.com/gb11TvBfn0
— Adrien Grondin (@adrgrondin) September 19, 2026
Memory Remains the Biggest Limitation
Despite the performance improvement, the iPhone 18 Pro still faces limitations when running larger AI models.
While the demonstration uses the original Bonsai model, the newer Bonsai 2 cannot comfortably run locally on the device when using 2-bit quantization because its memory requirements are too high. Attempting to fit the larger model into the iPhone’s available memory results in degraded performance.
The original Bonsai model uses 1-bit quantization, making it substantially more memory-efficient. It can run on devices with as little as 4GB of RAM, while phones with 8GB or 12GB have considerably more room to handle it.
This allows Bonsai 27B to take advantage of the iPhone 18 Pro’s faster memory and Neural Engine without the same memory pressure.
More RAM Could Unlock Larger Models
The demonstration highlights both the progress and remaining limitations of running large AI models directly on smartphones.
The A20 Pro and faster memory allow the iPhone 18 Pro to handle a dense 27B model much faster than its predecessor, but the phone’s 12GB memory configuration still prevents some larger or less aggressively quantized models from running efficiently.
Future iPhones with larger memory configurations could therefore support denser on-device AI models that remain impractical on the current hardware.
The post iPhone 18 Pro Can Run Massive 27B AI Model On-Device appeared first on ProPakistani.
