On this page
You can run Llama 3.2 on your Android phone with Off Grid AI Mobile (OGAM). Search for Llama 3.2, download the 1B (0.8 GB) or the 3B (2 GB), and you have Meta’s model running on your own phone with no internet, no account and no login.
It’s free, and the 1B runs on a 4 GB phone.
Google Play · Download Off Grid AI
What Llama 3.2 1B and 3B are
Meta released Llama 3.2 on 25 September 2024. The 1B and 3B are its text-only models, made for phones and other small devices. Meta’s model card lists eight languages it officially supports: English, German, French, Italian, Portuguese, Hindi, Spanish and Thai. The licence is the Llama 3.2 Community License.
Meta’s own repo is gated and doesn’t carry GGUF files, the format OGAM runs. Trusted quantizers such as bartowski and Unsloth publish open GGUF builds that need no login, and those are what OGAM finds.
Which size fits your phone
OGAM judges fit from your phone’s RAM. On Android, a download under 52% of RAM runs well, 52% to 62% is marked Tight, and anything bigger isn’t offered.
| Model | File (Q4_K_M) | File (Q8_0) | 4 GB phone | 6 GB phone | 8 GB+ phone |
|---|---|---|---|---|---|
| Llama 3.2 1B | 0.81 GB | 1.32 GB | Runs well | Runs well | Runs well |
| Llama 3.2 3B | 2.02 GB | 3.42 GB | Tight at Q4_K_M | Runs well at Q4_K_M | Runs well, Q8_0 too |
Sizes come from bartowski’s GGUF builds. Android reports a little less RAM than the box says, which is why the 3B lands on Tight on a 4 GB phone. To check your RAM, open Android Settings, then About phone.
Q4_K_M is the 4-bit build and the one to start with. Q8_0 is closer to the original and twice the size, so take it only when the table says it fits.
Set it up
- Install OGAM from Google Play.
- Tap Models in the bottom bar. You land on the Text tab.
- Type Llama 3.2 in Search models.
- Tap Llama-3.2-3B-Instruct-GGUF, or Llama-3.2-1B-Instruct-GGUF on an older phone. Each model shows once, led by its most trusted upload, with a +N uploads chip that counts the copies.
- Under Sizes, take the Recommended file and download it. Use Wi-Fi for the 3B.
- When it finishes, tap Chat now.
The cards say Newer: Qwen 3.5. OGAM is pointing you at the model it recommends today. It doesn’t stop you, and Llama 3.2 downloads and runs like any other model.
Once it’s downloaded, turn on airplane mode and send another message. It still answers, because the model runs on your phone.
What to ask it
A 3B model is good at short, everyday writing and tidying. Try:
- “Summarise this message in two lines.” Paste a long email or group chat.
- “Rewrite this so it sounds friendly but firm.” A reply you’d rather not write yourself.
- “Turn this into a shopping list by aisle.” Paste a recipe.
- “Translate this into Hindi.” Hindi is one of its eight official languages.
Don’t trust it for facts it has to remember. A 1B or 3B model gets dates, names and numbers wrong more often than a big cloud model, and its training data stops in 2023.
Llama 3.2 or Qwen 3.5?
Qwen 3.5 is newer, reads photos too, and is what OGAM lists first. Pick Llama 3.2 when you want Meta’s model in particular, or when you’re testing something built on Llama and want the same model on your phone. For the app’s own pick, see the Qwen 3.5 Android guide.
If something goes wrong
| Symptom | Check | Fix |
|---|---|---|
| The 3B isn’t offered, or only small files show | Your phone’s RAM | Use the 1B, or a smaller 3B file such as Q4_0 |
| The app closes when the 3B loads | It’s Tight on your phone, other apps hold memory | Close other apps, or switch to the 1B |
| Replies are short and generic | 1B models struggle with open questions | Give it the text to work on, or move up to the 3B |
| You see a login or licence screen | You opened Meta’s own page in a browser | You don’t need it. Download from OGAM’s search instead |
FAQ
Can Llama 3.2 run on an Android phone? Yes. Search for Llama 3.2 in OGAM and download a GGUF build of the 1B or 3B. It runs on the phone with no internet. You don’t need a Meta account or a Hugging Face login.
How much RAM does Llama 3.2 need on Android? The 1B’s 4-bit file is 0.81 GB, so it runs well on a 4 GB phone. The 3B’s is 2.02 GB, which is right at the line on a 4 GB phone and runs well on 6 GB and up.
Should I pick Llama 3.2 1B or 3B? The 3B if your phone has 6 GB of RAM or more. It is the stronger of the two. The 1B is for older phones or when you want the fastest replies.
Why does the app say Newer: Qwen 3.5 on Llama 3.2? Llama 3.2 came out in September 2024, and OGAM now recommends Qwen 3.5 for the same job. The label is a heads-up, not a block. Llama 3.2 still downloads and runs.
Is Llama 3.2 free to use? Yes. Meta releases it under the Llama 3.2 Community License, which allows personal and most commercial use, and OGAM is free with no account.
Related guides
- How to Run Local AI on Your Android Phone in 2026 (No Cloud, No Account)
- How to Run Qwen 3.5 on Your Android Phone in 2026 (Locally, No Cloud)
- How to Find AI Models Your Android Phone Can Actually Run in 2026 (No Failed Downloads)
- How to Fix a Local AI Model That Crashes on Android in 2026
Get Off Grid AI on Google Play and run Llama 3.2 on your own phone today.