Quick guide
What specs do you need for local AI?
For most people, the practical starting point is a small or standard local model: 3B, 4B, 7B, or 8B parameters in a quantized format. These are small enough to run on many modern laptops while still being useful for summaries, planning, notes, and everyday chat.
GPU memory matters most on Windows and Linux. A dedicated GPU with 6GB to 8GB VRAM can usually handle smaller local models, while 12GB to 16GB opens up more comfortable 7B, 8B, and some 14B workflows. CPU-only can work, but it often feels slower.
On Apple Silicon Macs, unified memory is the important number. An 8GB Mac can try small models, 16GB is a stronger baseline, and 32GB+ gives you more room for better models, longer context, and normal apps running at the same time.