You find a model on Hugging Face, wait for a 5 GB download on your iPhone, and then it will not load: too big for the phone's memory, or a format the app cannot run yet. Off Grid AI Mobile (OGAM) now removes those results before you tap download. Model search on your iPhone lists only models it can load, shows each model once instead of a page of copies, and marks the files that only just fit.
The app uses one rule everywhere it offers a model: the picks, the size list on a model's page, and the warning before a load.
iOS does not give one app all of the phone's memory, and a running model needs more than its file size: the weights, plus the cache that holds your conversation. The app's phone budget comes from a measurement on a 12 GB iPhone 17 Pro Max: Gemma 4 E4B, a 4.64 GB download, needed 6.4 GB at a 4K context, about 1.4 times its file size. From that, in OGAM 0.0.117:
| iPhone RAM | Runs well | Tight | Hidden |
|---|---|---|---|
| 6 GB | under 2.2 GB | 2.2 to 3 GB | over 3 GB |
| 8 GB | under 2.9 GB | 2.9 to 4 GB | over 4 GB |
| 12 GB | under 4.3 GB | 4.3 to 6 GB | over 6 GB |
The app uses the RAM iOS reports, which is a little under the number on the box. A 12 GB iPhone 17 Pro Max reports 11.2 GB, so its limits sit slightly below the last row.
Hugging Face search results do not carry file sizes, so the list judges them by parameter count at 4-bit, about 0.6 GB per billion parameters. A 7B model works out at about 4.2 GB, so it does not appear on an 8 GB iPhone. A 9B model, at about 5.4 GB, does appear on a 12 GB iPhone.
Size is not the only check. Some new models need a newer llama.cpp engine than the one inside your version of the app, and every upload of those models is left out, so you do not download something that fails to load. Files that only speed up another model (draft and MTP files) are not offered as models. Image generators stay on the Image tab and do not show in chat search.
qwen, gemma or lfm. Results appear half a second after you stop typing, or when you tap search on the keyboard.
Matching models from the app's own catalog come first. Hugging Face results follow, up to 30 per search, and a model already in the curated list is not repeated.
Each result is one model, not one upload. When several people uploaded the same model, you see it once, led by the most trusted upload, with a +N uploads chip that counts the others.
Beyond the name, maker and size, a card can show:
Abliterated and uncensored uploads are not shown.
Next to the search field are three icons.
The third icon, Import local file, opens the Files picker if you already have a GGUF file on your iPhone.
Why can't I find a model on my iPhone that I saw on Hugging Face?
It is over half of your iPhone's RAM, it needs a newer engine than your app version has, or it is an image generator, which lives on the Image tab. Abliterated and uncensored uploads are also hidden.
What is the biggest AI model an iPhone can run?
In OGAM 0.0.117, search offers files up to half of the RAM iOS reports: about 3 GB on a 6 GB iPhone, 4 GB on an 8 GB iPhone and just under 6 GB on a 12 GB iPhone 17 Pro Max. At 4-bit, that is up to about 4B, 6B and 9B parameters.
What does Tight mean in Off Grid AI on iPhone?
The file is between 36% and 50% of your iPhone's RAM. It loads, but there is little memory left, so a long conversation or switching apps can cause trouble. A smaller size is steadier.
What does "3B active" mean on an MoE model?
It is a mixture-of-experts model that uses 3 billion parameters per token, so it replies at about the speed of a 3B model. It still needs memory for all of its parameters.
Why does a model show "+3 uploads"?
Three other people uploaded the same model. The app shows the most trusted upload and counts the rest, so the list does not fill with copies.
Can I search for Liquid AI models on iPhone?
Yes. Open the filters, tap Org and pick Liquid AI, or type lfm in the search field.