Which AI model will run on my computer?

Every model download page is full of numbers: 8B, Q4_K_M, 32K context. Here's what actually decides whether a model runs well on your computer.

Memory is the first limit

A model has to fit in memory to run. Roughly, a model needs about 0.6 GB per billion parameters in its usual compact (4-bit) form, plus room for the conversation:

Model size	Download	Memory to run comfortably
4B	~2.5 GB	8 GB computer
8B	~5 GB	16 GB computer
14B	~9 GB	16 to 24 GB computer
30B+	18 GB and up	32 GB or more

On a PC with a graphics card, the model runs fastest when it fits in the card's own memory (VRAM). On a Mac, memory is shared, so most of it can be used by the model.

Speed comes from memory bandwidth

Each word the model writes reads through the whole model. So speed depends on how fast memory can be read: a graphics card is many times faster than ordinary RAM, and Apple's Pro and Max chips are faster than the base ones.

Mixture-of-experts models are the exception that helps: a "35B-A3B" model has 35B parameters but uses only about 3B for each word, so it writes quickly while needing the memory of a big model.

Let the app do the arithmetic

You don't have to work any of this out. Zero1LLM reads your hardware and labels every model Runs well, Runs slowly or Won't fit, then picks the version that suits your computer. On this site, the model catalog uses the same rules: open a model and pick a computer like yours under "Will it run?".