AI on your iPhone used to mean sending your words to a server. In 2026 your phone runs real models by itself, with airplane mode on.
See what your iPhone can run
| Your iPhone | What runs offline |
|---|---|
| Runs Apple Intelligence | Apple's on-device model, built into iOS |
| 8 GB of memory | Open models up to about 4B parameters, like Qwen 3.5 4B |
| 6 GB of memory | Open models around 2 to 3B, like Granite 4.2 3B |
| Older iPhones | Small models only, slowly |
Every iPhone 16 and newer has 8 GB of memory, and so do the iPhone 15 Pro and Pro Max.
Pick the kind of app you need
Offline chat apps like Locally AI, PocketPal and Enclave give you a chat box with a local model. Use them to draft, brainstorm or ask general questions on a plane.
Offline AI for your own files reads what you already have. Vaulteo uses local models to file your papers, pull out amounts and due dates, and answer questions about your documents and photos inside an encrypted vault.
Know how fast it answers
On an iPhone 16 Pro, Qwen 3.5 4B read one of our sample papers and filled every field in about 19 seconds. Apple's on-device model answers short questions within a few seconds. Older and smaller phones take longer.
Prove an app stays offline
- Turn on airplane mode, and turn Wi-Fi off too.
- Open the app and ask it something new.
- If you get an answer, the model runs on your iPhone.
Then check the app's privacy label on the App Store. "Data Not Collected" plus a working answer in airplane mode tells you most of what you need.
Know the limits
- Local models know less about the world than the biggest cloud models.
- They make mistakes. Check names, numbers and dates against the source.
- A model download takes 2 to 3 GB of storage.
For work on your own private files, those trade-offs buy you something no cloud can: your papers never leave your hand.