On this page
You can run LFM3 on your Android phone with Off Grid AI Mobile (OGAM). Download one model of about 1 GB once, and you get a fast chat model, or one that reads your photos, running on your own phone with no internet and nothing uploaded.
It’s free, and it fits phones with 4 GB of RAM, so you don’t need a flagship to try it.
Google Play · Download Off Grid AI · App Store
What LFM3 is
LFM3 is Liquid AI’s newest small model family, released in October 2026. It mixes short convolution layers with attention layers, a design Liquid made to run on phones and laptops. Liquid says the 1.7B model takes about 1 GB of memory at 4-bit, which is the version OGAM downloads.
OGAM 0.0.122 added two LFM3 models to its catalog:
| Model | What it reads | Shows in the app as | RAM listed |
|---|---|---|---|
| LFM3 1.7B | Text | 1.0 GB | 4 GB or more |
| LFM3 VL 2B | Text and photos | 1.4 GB with its vision file | 4 GB or more |
Both download at Q4_K_M. The VL model comes as two files, the model and a 361 MB vision file (an “mmproj” file at F16). OGAM downloads them together and pairs them when the model loads, so you don’t hunt for the second file on Hugging Face.
To check your RAM, open Android Settings, then About phone.
What LFM3 1.7B is good for
OGAM’s catalog describes LFM3 1.7B as “Fast agent and tool use with reasoning; text only”. Liquid trained it for tool calling and structured answers, and it understands 16 languages, including Hindi, Arabic, Japanese, Spanish and Russian. Here’s what to ask it:
- “Turn these notes into a packing list, grouped by bag.” Paste messy text and get it back in order.
- “Reply to this message politely and say I can do Thursday instead.” A short draft you can copy straight into WhatsApp.
- “Put these expenses in a table with a total.” Structured output is what it was trained for.
- “Rewrite this in simple Hindi.” Hindi is one of the 16 languages it was trained on.
Liquid says the 1.7B model isn’t meant for long coding jobs or hard multi-step reasoning. If you want a stronger all-rounder and have 8 GB of RAM or more, the Qwen 3.5 guide covers the next size up. LFM3 is the one to pick when you want a quick answer on a phone with less memory.
What LFM3 VL 2B lets you do with photos
LFM3 VL 2B is the LFM3 language model with a 164M-parameter vision encoder added. The catalog line is “Fast reasoning with vision, reads documents and charts”, and Liquid lists reading documents, charts and app screens as what it’s built for. Take a photo or pick a screenshot, then ask:
- A printed bill or receipt: “What’s the total, and how much of it is tax?”
- A screenshot of a chart: “Which month had the highest number?”
- A page of a letter or form: “What’s the deadline in this letter?”
- A photo of a shelf or a desk: “Where is the blue folder?” Liquid trained it to find objects and say where they sit in the picture.
The photo stays on your phone. A cloud assistant has to upload your bill or your letter to answer, and OGAM doesn’t.
Set it up
- Install OGAM from Google Play. You need version 0.0.122 or later.
- Tap Models in the bottom bar. You land on the Text tab.
- Type LFM3 in Search models. The catalog’s LFM3 models come first, above any Hugging Face results.
- Tap LFM3 1.7B for chat, or LFM3 VL 2B if you want to ask about photos, and download it. Use Wi-Fi, because it’s about 1 to 1.4 GB.
- Go to Home and tap Select Model, then pick the LFM3 model you downloaded.
- Tap New Chat and type your question.
To ask about a photo with LFM3 VL 2B, tap + in the message box, choose Photo, then Camera or Photo Library. Add your question and send. If Photo is greyed out and the app says “Vision Not Supported”, LFM3 1.7B is still the active model, so switch to LFM3 VL 2B.
Once it’s downloaded, turn on airplane mode and send another message. It still answers, because nothing leaves the phone.
If something goes wrong
LFM3 doesn’t show up in the list. Update OGAM to 0.0.122 or later from Google Play. The LFM3 models came with that release, and they need the newer llama.cpp engine it ships.
The app says “Vision File Missing”. The model is there but its vision file isn’t. Tap Go to Download Manager, then the wrench next to LFM3 VL 2B to download it, and reload the model.
FAQ
Can LFM3 run on an Android phone? Yes. Off Grid AI Mobile has LFM3 1.7B and LFM3 VL 2B in its model catalog. Both are listed for phones with 4 GB of RAM or more, and both run on the phone with no internet.
How big is the LFM3 download? LFM3 1.7B shows as 1.0 GB in the app. LFM3 VL 2B is the same size plus a 361 MB vision file, so about 1.4 GB in all. The app fetches both files together.
What is the difference between LFM3 1.7B and LFM3 VL 2B? LFM3 1.7B reads text only. LFM3 VL 2B is the same language model with a vision encoder added, so you can attach a photo and ask about it. Pick VL if you want to ask about pictures.
Is LFM3 free to use on Android? Yes. Chat and photo questions are part of the free Off Grid AI app. There’s no account, no subscription and no limit on messages.
Does LFM3 work offline? Yes, after the first download. The model runs in your phone’s memory, so you can turn on airplane mode and keep chatting. Your messages and photos are never uploaded.
Related guides
- How to Run Qwen 3.5 on Your Android Phone in 2026 (Locally, No Cloud)
- How to Run Vision AI Locally on Your Android Phone in 2026 (No Cloud, No Subscription)
- How to Find AI Models Your Android Phone Can Actually Run in 2026 (No Failed Downloads)
- How to Ask AI About a Screenshot on Android in 2026 Without Internet
Get Off Grid AI on Google Play and run LFM3 on your own phone today.