On this page
When an AI model won’t load on your iPhone, or the app closes as it loads, the model needs more memory than iOS will give one app. A running model needs its file plus room for the conversation. Pick a file under about 60% of your iPhone’s RAM, lower the context length, close other apps, and unload any other model. If the model you want is bigger than that, run it on your own computer and chat from your iPhone.
The steps use Off Grid AI Mobile (OGAM), the free iPhone app, version 0.0.119 or later. The same limits apply to every local AI app on iPhone.
Why your iPhone runs out of memory
iOS doesn’t give one app all of the phone’s RAM, and it closes apps that go over their share. OGAM measured real loads on a 12 GB iPhone 17 Pro Max: a Qwen 3.5 9B build (5.6 GB) and Gemma 4 E4B (5.0 GB) both load and answer at a 4K context. From that, OGAM sizes models against your iPhone’s RAM:
| iPhone RAM | Runs well | Tight | Too big |
|---|---|---|---|
| 6 GB | under 3.6 GB | 3.6 to 4.2 GB | over 4.2 GB |
| 8 GB | under 4.8 GB | 4.8 to 5.6 GB | over 5.6 GB |
| 12 GB | under 7.2 GB | 7.2 to 8.4 GB | over 8.4 GB |
iOS reports a little less RAM than the box says, so the real limits sit slightly lower. At 4-bit, a model is about 0.6 GB per billion parameters. That puts a 7B model at about 4.2 GB, which runs well on an 8 GB iPhone, and a 13B model at about 7.8 GB, which only a 12 GB iPhone can load.
Fix it in order
1. Pick a file that fits. Open OGAM, tap Models and search. Search only lists files your iPhone can load. On a model’s page, Sizes shows Recommended, Smaller and Best. A file marked Tight loads but leaves little room for a long chat or other apps. If loads fail, take Smaller, or the next model size down.
2. Close other apps. OGAM’s fit check uses total RAM, not what’s free right now. If other apps hold a lot of memory, a Tight model can still fail. Close the camera, games and large browser sessions, then load again. When a load fails, OGAM keeps your current model loaded.
3. Unload the other model. Open the model selector in a chat. Currently Loaded shows what’s in memory. Tap Unload on anything you’re done with. Unloading frees memory without deleting the file.
4. Lower the context length. Go to Settings → Model Settings and open the text generation controls. Context Length sets how much conversation the model holds, and a bigger context needs more memory. Lower it, reload, and start a fresh chat for a new topic.
5. Use the tuned settings. Under Advanced, Use fastest settings puts the model on the GPU with Metal and sets the cache to q8_0, which uses about half the memory of f16. If you changed settings by hand and loads started failing, this puts you back on known-good values.
Symptom, check, fix
| Symptom | Check | Fix |
|---|---|---|
| Model won’t load | File size vs your iPhone’s row above | Smaller size or a smaller model |
| Loads, then the app closes in a long chat | Context Length | Lower it, start a new chat |
| A Tight model fails sometimes | Other apps open | Close them and load again |
| Fails after loading a second model | Currently Loaded | Unload the first one |
| Model fits but is slow | Older chip | Use fastest settings, or a smaller model |
Run the bigger model from your own computer
If the model is too big for any iPhone, let your computer hold it. Install Off Grid AI Desktop on your Mac or PC, load the model there, then in OGAM open Settings → Remote servers. Add the computer’s address on port 7878 (for example http://192.168.1.50:7878), tap Test connection, and pick the model. Allow Local Network access for OGAM when iOS asks. The computer runs the model and your iPhone shows the answer. Keep it on your home network, and don’t forward port 7878 to the internet.
Need the full-size model?
Some models are too big for your iPhone and your computer. Off Grid AI Boost is an API for open models on our GPUs, at one flat price from $4.99 a month for the first cohort (regular price $9.99, plans up to $199.99). It names Qwen 3.8 and DeepSeek V4.1 Flash, gives you one API key for every model, and we don’t store your prompts or train on them. The full privacy policy comes with launch.
It’s a pre-order. Boost starts in January; pre-order to lock in half price. Usage limits for each plan aren’t published yet. Keep a small model on your iPhone for offline, private use, and use Boost for the big one.
FAQ
How much memory can an iPhone app use for AI? Less than the phone’s total RAM, and Apple doesn’t publish one number. OGAM’s rule, from measured loads, is that files under 60% of your RAM run well and files over 70% are marked Too big.
What’s the biggest model an iPhone can run? In OGAM, about 4.2 GB files on a 6 GB iPhone, 5.6 GB on 8 GB, and about 7.8 GB on a 12 GB iPhone 17 Pro Max (it reports 11.2 GB). At 4-bit that’s roughly 7B, 9B and 13B parameters.
Why does the model load sometimes and fail other times? The fit check uses total RAM, but what’s free changes with the apps you have open. Close apps or pick the Smaller size.
Does a smaller context make answers worse? The model sees less of the conversation at once. For a single question or a short rewrite you won’t notice.
Can I run a 13B or 30B model on my iPhone? Not on the iPhone itself. Run it on your computer with Off Grid AI Desktop and connect from OGAM, or use Boost once it starts.
Open Models, tap the model that failed, and download its Recommended or Smaller size. That fixes most “not enough memory” loads.