localAI Hardware Calculator A hardware calculator for local AI. Detect your Mac, GPU or laptop, pick what you want to do, and see which open models run well on it.

localAI Hardware Calculator

What AI can I run on my machine?

About localAI Hardware Calculator

Tell us your hardware and what you want to do, and get a list of open models that run comfortably on it. Everything runs in your browser and nothing is sent anywhere.

How the estimate works
  • Memory first. A model has to fit in memory to run well. Most text models are downloaded at 4-bit, which takes about 0.6 GB per billion parameters, plus about 1.5 GB for the conversation and the app running it.
  • Macs share their memory. The GPU on Apple Silicon can use about two thirds of memory up to 36 GB, and three quarters above that. The rest is left for macOS and your apps.
  • Speed comes from memory bandwidth. To write each token (about three quarters of a word), the computer reads every active weight once. So tokens per second is roughly bandwidth divided by model size. That's why a big Mac or a fast GPU matters more than the CPU.
  • Mixture-of-experts models are fast for their size. A model like Qwen3.6 35B A3B needs memory for all 35B parameters, but only reads 3B of them per token.
  • PCs can spill over. A model too big for the graphics card can run partly from system RAM, but that part runs at RAM speed, which is much slower.
  • Image models are different. They need room for the image model plus its text encoder, and their speed depends on how many steps they take, so we show steps instead of tokens per second.

These are estimates, so treat them as a starting point. Real speed depends on the app (Ollama, LM Studio, MLX, llama.cpp), how long your conversation is, and what else is running.

Published Oct 29th, 2026 · Updated Nov 4th, 2026