Yes, many older phones can run AI locally. If your phone runs iOS 15.1 or later, or Android 7 or later, and has about 3 GB of RAM, it can run a small language model that chats, rewrites text and answers questions with no internet. With 2 GB you can still run a very small model. What you can’t get on an older phone is Apple Intelligence or Gemini Nano, because Apple and Google limit those to recent hardware.
Expect slower answers and simpler reasoning than on a new flagship. For everyday tasks, a small model on an old phone is still useful.
Why can’t older phones get Apple Intelligence or Gemini Nano? #
Apple Intelligence requires an iPhone 15 Pro or newer, which means an A17 Pro chip or later and 8 GB of RAM. Every iPhone before that, including the regular iPhone 15, is left out.
On Android, Gemini Nano runs only on phones Google has certified for it, such as the Pixel 9 and later, the Galaxy S25 and S26 series and a list of recent flagships from other brands. A three-year-old mid-range phone won’t get it.
Third-party local AI apps don’t have that restriction. They run open models through llama.cpp, which works on the GPU where it can and on the CPU everywhere else. If the model fits in memory, it runs.
Which models fit a phone with 2, 3 or 4 GB of RAM? #
Here’s what the Personal LLM catalog needs, which is a fair guide for local AI in general.
| Your RAM | What runs | Good first pick |
|---|---|---|
| 2 GB | Qwen 3.5 0.8B | Qwen 3.5 0.8B (0.81 GB) |
| 3 GB | Gemma 4 E2B, Ministral 3 3B, Qwen 3.5 4B | Gemma 4 E2B (2.04 GB) |
| 4 GB | The same models with more headroom | Qwen 3.5 4B (2.74 GB) |
| 5 to 6 GB | Adds Gemma 4 E4B, and Qwen 3.5 9B at 6 GB | Qwen 3.5 4B or Gemma 4 E4B |
| 8 GB+ | Everything, including GLM 4.6V Flash | Qwen 3.5 9B |
The minimums are just that. A model “should run” when your phone meets its minimum, but it runs more comfortably with room to spare. Personal LLM marks a model “Fits your device” only when your phone has 2 GB more than the minimum, and it reads your real RAM, so you don’t have to find it yourself.
How do you find out how much RAM your phone has? #
- iPhone: Settings doesn’t show RAM. As a rough guide, iPhone 13 has 4 GB and iPhone 14 and 15 have 6 GB. Older models generally have less. Check your exact model’s spec sheet, or install a local AI app that reports it.
- Android: Look in Settings > About phone. Some brands show RAM there, others list it under Memory or in Developer options. The spec page for your model will have it too.
What can a small model actually do? #
A 0.8B model is fast but limited. It can rephrase a sentence, write a short reply, give a simple definition and hold a light conversation. It makes more mistakes, loses track in long chats, and is weak at math and facts. Treat it as a pocket helper for text, not a source of truth.
A 2B to 4B model is a real step up. Gemma 4 E2B was designed by Google DeepMind for phones, and Qwen 3.5 4B is the best balance of speed and quality in the Personal LLM catalog. Both handle:
- Drafting and rewriting emails and messages
- Summarizing an article or notes you paste in
- Explaining concepts and answering general questions
- Language practice and translation help
- Brainstorming lists, plans and ideas
They still struggle with long reasoning chains, niche facts and very long documents. Why AI makes things up covers how to catch the mistakes.
How fast will it be on an older phone? #
Slower than on a new one, and the difference depends on the chip more than the RAM.
On iPhone, local AI apps use the GPU through Apple’s Metal framework, so even older iPhones get acceleration. On Android, Personal LLM uses the GPU through OpenCL on Snapdragon phones with Adreno 700-series graphics or newer, which means recent Snapdragon chips. Other Android phones fall back to the CPU, which is noticeably slower.
Rough expectations:
- A 0.8B model feels quick on almost anything.
- A 2B to 4B model on an older phone may produce text a little slower than you read. That’s usable for short answers and frustrating for long ones.
- The first reply after opening the app takes longer, because the model has to load into memory.
Personal LLM shows tokens per second under every reply and has a built-in benchmark, so you can compare models on your phone instead of guessing.
How to get the most out of an old phone #
- Start small. Try the smallest model that fits, then move up if speed is fine.
- Keep thinking mode off. It makes the model write out its reasoning first, which multiplies the wait. Turn it on only for math or logic.
- Start new chats often. Long conversations fill the context window, which slows things down and uses memory.
- Lower the context size and max tokens in settings if the app lets you. Shorter limits mean less memory and faster answers.
- Close other apps before a long session, especially games, camera apps and browsers with many tabs.
- Keep the phone cool. Older phones throttle when hot. Take it out of a thick case and don’t run AI while fast charging.
- Check storage. Old phones are often nearly full. See how much storage AI models need.
For more speed fixes, read why local AI is slow and how to speed it up.
Is it worth running AI on an old phone? #
If you want private, offline help with everyday writing and questions, yes. An old phone that sits in a drawer can become a dedicated offline assistant for travel or for sensitive notes, and the models are free.
It isn’t worth it if you need strong reasoning, long document analysis or fast, long answers. For that, a newer phone with 8 GB or more, or a cloud chatbot, will serve you better.
Frequently asked questions #
Can a phone with 2 GB of RAM run AI? #
Yes, but only a very small model. Qwen 3.5 0.8B needs about 2 GB and is built for low-end phones. It’s fast, but its answers are basic and less reliable than larger models.
Does my iPhone 13 support AI? #
It doesn’t support Apple Intelligence, which needs an iPhone 15 Pro or later. It can run local AI apps. With 4 GB of RAM, an iPhone 13 can run models like Gemma 4 E2B, Ministral 3 3B and Qwen 3.5 4B.
What Android version do I need for local AI? #
It depends on the app. Personal LLM supports Android 7 and later. Google’s AI Edge Gallery needs Android 12 or later. RAM and chip speed matter more than the Android version for how well it runs.
Will running AI damage an old phone? #
No. It works the phone hard, like a demanding game, so it gets warm and uses battery faster. Heat is hard on batteries over time, so let the phone cool if it gets hot and avoid long sessions while charging. Our post on AI and battery drain has details.
Why is AI so slow on my old Android phone? #
Most likely it’s running on the CPU because the phone’s GPU isn’t supported. Use a smaller model, keep thinking mode off, and start new chats often.