AI on Android: The Ultimate Compatibility List
So you want to run AI on your old phone? Most models (like the new GPT-OSS 20B) are too big and will crash your device.
Here is the list of models that actually work, categorized by how much RAM your phone has.
🟢 Tier 1: The "Potato" Phone (3GB - 4GB RAM)
These models are tiny, fast, and will run on almost anything.
1. Qwen 2.5 (0.5B) The speed king. It's not a genius, but it replies instantly.
ollama run qwen2.5:0.5b
2. Llama 3.2 (1B) Meta's optimized mobile model. Surprisingly smart for its size.
ollama run llama3.2:1b
3. TinyLlama (1.1B) The classic. Good for basic chat and testing.
ollama run tinyllama
🟡 Tier 2: The Sweet Spot (6GB RAM)
This is where you get real intelligence. These can code, summarize, and reason.
1. DeepSeek Coder (1.3B) The best for coding questions. If you want your phone to write Python, use this.
ollama run deepseek-coder:1.3b
2. Qwen 2.5 (1.5B) A perfect balance of speed and smarts. Great for general questions.
ollama run qwen2.5:1.5b
3. Phi-3.5 Mini (3.8B) – Warning: Heavy Microsoft's model. It's very smart but might be slow on older chips. Give it a try!
ollama run phi3.5
🔴 Tier 3: The Danger Zone (8GB+ RAM Only)
Only run these if you have a newer phone or a gaming phone.
1. Llama 3.2 (3B) The standard for mobile AI right now.
ollama run llama3.2
2. Mistral (7B) - Heavily Quantized You might be able to squeeze this in if you close all background apps.
ollama run mistral
Quick Troubleshooting
- "Killed" Error: Your phone ran out of RAM. Try a smaller model from Tier 1.
- Slow Speed: This is normal on old CPUs. Be patient!
- Heat: Your phone will get hot. That means it's working.