IPhone 17 Excluded From iOS 27’s New AI Dictation System
What You Need to Know
- Apple’s new AI dictation system requires 12GB RAM, excluding standard iPhone 17 from the feature.
- AFM 3 Core Advanced model preferred over existing dictation by 44.7% to 17.6% in human evaluations.
- Apple stores 20-billion-parameter model in flash memory, activating only 1-4 billion parameters per request.
- New dictation feature available only on iPhone 17 Pro, Pro Max, Air, and select iPad/Mac models.
The most interesting angle buried in this piece is not that the feature exists, but that the standard iPhone 17 is quietly excluded from it despite being a current flagship. That hardware split tells a more interesting story than the AI accuracy numbers.
Apple’s new AI dictation system, arriving in iOS 27, runs on a 20-billion-parameter model called AFM 3 Core Advanced. The catch: it requires at least 12GB of RAM, and that 12GB requirement draws a line that leaves the standard iPhone 17 on the wrong side of it.
The standard iPhone 17 ships with 8GB of RAM, which means the upgraded dictation is reserved for the iPhone 17 Pro, iPhone 17 Pro Max, iPhone Air, Vision Pro with M5, iPads with M4 or later, and Macs with M3 or later. The base model iPhone 17 is excluded entirely, which is a quiet but meaningful product tier distinction Apple has not made loudly.
The accuracy case for the new system is straightforward. In human evaluations across seven quality dimensions including punctuation, casing, disfluency handling, and meaning capture, AFM 3 Core Advanced was preferred over Apple’s existing dictation system by a margin of 44.7% to 17.6%.
How the model fits on a phone
To run a 20-billion-parameter model on a smartphone, Apple stores the full model in flash memory rather than DRAM. A lightweight routing block selects a fixed set of “experts” during processing and reselects them periodically, a technique Apple calls Instruction-Following Pruning. The model activates only one to four billion parameters at a time depending on the request.
The feature is off by default in the first developer beta, and the same model also powers Apple’s new customizable Siri voices, also opt-in. Whether Apple flips dictation on automatically before the public release remains unclear, though the EU rollout pattern from iOS 18 suggests staged availability is the more likely path. The feature runs entirely on-device, so network connectivity has no effect on transcription quality.
0 Comments