StoweEvery iPhone Apple has allowed to run Apple Intelligence has at least 8GB of RAM, and every iPhone it has excluded has less. That is the whole rule, and it is a better predictor than chip generation, marketing tier, or price. This page is the reference version: which models have how much memory, why the line sits where it does, and what each tier can actually run — Apple's own models and third-party ones.
One line of disclosure: I build Stowe, a planner whose AI runs on-device, and this page exists because "which phones can actually do that" is the question behind every other question I get. Nothing below depends on the app.
Which iPhones have how much RAM?
Apple does not publish RAM figures for iPhones — the numbers below come from teardown-based databases (Everymac, AppleDB) and, for the 17 series, from identifiers Apple itself shipped in Xcode, as reported by MacRumors.
| Model (year) | Chip | RAM | Apple Intelligence |
|---|---|---|---|
| iPhone 12 / 12 mini (2020) | A14 | 4GB | No |
| iPhone 12 Pro / Pro Max (2020) | A14 | 6GB | No |
| iPhone 13 / 13 mini (2021) | A15 | 4GB | No |
| iPhone 13 Pro / Pro Max (2021) | A15 | 6GB | No |
| iPhone 14 / Plus (2022) | A15 | 6GB | No |
| iPhone 14 Pro / Pro Max (2022) | A16 | 6GB | No |
| iPhone 15 / Plus (2023) | A16 | 6GB | No |
| iPhone 15 Pro / Pro Max (2023) | A17 Pro | 8GB | Yes — the first supported iPhones |
| iPhone 16 / Plus (2024) | A18 | 8GB | Yes |
| iPhone 16 Pro / Pro Max (2024) | A18 Pro | 8GB | Yes |
| iPhone 16e (2025) | A18 | 8GB | Yes — the cheapest supported iPhone |
| iPhone 17 (2025) | A19 | 8GB | Yes |
| iPhone Air (2025) | A19 Pro | 12GB | Yes, incl. the 12GB-only tier |
| iPhone 17 Pro / Pro Max (2025) | A19 Pro | 12GB | Yes, incl. the 12GB-only tier |
Read the Apple Intelligence column against the RAM column and the pattern is exact: 8GB yes, less than 8GB no, with no exceptions in either direction. The base iPhone 15 shipped ten months before Apple Intelligence was announced, has the same A16 chip as the 14 Pro, and is excluded; the budget 16e is the cheapest phone in the lineup and is included. Memory, not marketing tier, draws the line. Apple's own supported-device list matches this table.
Why is 8GB the line?
Two numbers collide: the size of a useful language model, and the fraction of a phone's RAM an app is actually allowed to use.
Apple's on-device foundation model is approximately 3 billion parameters, per Apple's own machine-learning research publications. Quantized for deployment, a model that size occupies roughly 2GB of memory before you count the working state (the KV cache) that grows with the text being processed. That model has to be resident in RAM while iOS itself, the system's other services, and the app you are actually using are all also resident.
Meanwhile iOS enforces hard per-process memory ceilings — cross them and the system kills the process, a mechanism Apple documents as jetsam. The ceilings are not published, but developer measurements put a foreground app's limit at roughly half of installed RAM — for example, about 2.1GB on a 4GB iPhone 12, with an entitlement available that raises the cap somewhat on newer devices. On a 6GB phone, holding a ~2GB model resident means either evicting most other apps and system state or being killed; on an 8GB phone the same model fits with room for the rest of the system to breathe. That is the 8GB rule from the engineering side: it is not that a 6GB phone cannot execute the math, it is that the model does not fit in the memory an app or system service is allowed to keep.
What changed at 12GB?
In 2026 Apple split its on-device lineup in two: iOS 27's most capable on-device model requires 12GB of RAM, as reported by MacRumors — which today means only the iPhone Air, 17 Pro and 17 Pro Max qualify. The standard iPhone 17, with 8GB, runs the smaller on-device model, and features that want the larger one fall back to Private Cloud Compute. The same mechanics as 2023's 8GB line, one shelf higher: a bigger model needs a bigger resident budget, and only the 12GB devices have it.
The takeaway for anyone buying a phone for local AI in 2026: 8GB is the entry ticket, 12GB is the comfortable tier, and the gap between them is now a product line, not a rounding error.
What about iPads and Macs?
The same rule, dressed differently. Apple Intelligence on iPad requires an M-series iPad (Pro or Air) or the iPad mini with A17 Pro — every one an 8GB or higher configuration — while the base iPad (A16, 6GB) is excluded. On the Mac, every Apple silicon machine qualifies, because no M-series Mac ever shipped with less than 8GB of unified memory. The floor is the same floor.
What model sizes fit in each tier?
For third-party local models (the llama.cpp / MLX ecosystem), the arithmetic is simple enough to do on a napkin: at 4-bit quantization — the standard quality/size trade for phones — model weights cost roughly 0.6GB per billion parameters, and you budget extra for the KV cache and the app around it. Against the roughly-half-of-RAM foreground ceiling:
| RAM tier | Example devices | Practical model budget | Fits at 4-bit |
|---|---|---|---|
| 4GB | iPhone 12, 13 | ~1GB | ≤1B params — toy tier |
| 6GB | iPhone 14, base 15 | ~1.5GB | 1–2B params — small assistants, no Apple Intelligence |
| 8GB | 15 Pro, 16 series, 16e, 17 | ~2.5GB | 3–4B params — Apple Intelligence tier; where local AI gets useful |
| 12GB | iPhone Air, 17 Pro | ~4–5GB | 7–8B params — small-desktop-class models |
The budgets are deliberately conservative — a model that fits with nothing to spare will be killed the first time the camera or a heavy website wants memory. The qualitative jump is between the 6GB and 8GB rows: 1–2B models can classify and autocomplete, but the 3–4B class is where a local model starts reliably understanding loose, messy human language — which is exactly the work Apple sized its own ~3B foundation model for. That is why the 8GB rule is really a usefulness rule wearing a hardware spec.
If you keep one sentence from this page: check the RAM, not the chip. The A-series number tells you how fast the AI would run; the memory tells you whether it runs at all.
Frequently asked questions
Can the iPhone 15 run Apple Intelligence?
The base iPhone 15 and 15 Plus cannot — they have 6GB of RAM and the A16 chip, below Apple’s floor. The iPhone 15 Pro and Pro Max can: they were the first iPhones with 8GB of RAM, and in 2023 that made them the only supported models at launch.
Why does Apple Intelligence need 8GB of RAM?
Because the on-device foundation model (about 3 billion parameters) has to fit in memory alongside iOS and the app you are actually using, and iOS only lets a foreground app and system services claim a fraction of installed RAM. On a 6GB phone there is not enough headroom to hold the model resident without evicting everything else; 8GB is the smallest configuration where there is.
How do I find out how much RAM my iPhone has?
Apple does not list RAM on its spec pages — it never has for iPhone. Look the model up on a teardown-based spec database such as Everymac or AppleDB, or check the table in this post, which covers every model from the iPhone 12 forward.
Does more RAM make Apple Intelligence smarter?
Since 2026, yes. Apple’s most capable on-device model in iOS 27 requires 12GB of RAM, which only the iPhone Air, 17 Pro and 17 Pro Max have; 8GB devices run the smaller model or route those features through Private Cloud Compute. More RAM also raises the ceiling for third-party local models — roughly 7–8B parameters at 4-bit versus 3–4B on an 8GB phone.
Can older iPhones run any local AI at all?
Some, modestly. A 6GB iPhone can run a 1–2B parameter model at 4-bit quantization through apps that bundle their own runtime, and on-device speech recognition works back to much older hardware. What older phones cannot do is run Apple Intelligence or the 3–4B class of models where local AI starts being genuinely useful for understanding loose language.