Local AI hardware
What local AI models can my computer run?
Enter your Mac or PC specs and get a plain-English estimate of the model sizes that should feel smooth, usable, slow, or unrealistic.
Good first target
Most modern 16GB laptops should start with an 8B Q4 model.
Bigger can work, but speed and memory headroom matter more than the model name.
3B class
Q4
Phi, Gemma, Qwen small models
Fast drafts, simple chat, light extraction, small helper bots.
Comfortable
9GB headroom
7B class
Q4
Mistral 7B, CodeLlama 7B, similar small open models
Everyday chat, summarizing, basic coding help, lightweight private assistants.
Usable
7GB headroom
8B class
Q4
Llama 3.1 8B, Qwen 8B, similar mainstream local models
Best beginner target for most laptops: chat, writing, research, and moderate coding.
Usable
6GB headroom
14B class
Q4
Qwen 14B, Gemma larger local models
Stronger reasoning than 7B/8B, usually slower on laptops.
Usable
2GB headroom
32B class
Q4
Qwen 32B, DeepSeek Coder 33B style models
Serious local coding and analysis on high-memory Macs or GPU workstations.
Not recommended
10GB short
70B class
Q4
Llama 70B, Qwen 72B, large frontier-adjacent open models
High quality local inference, but hardware demands are workstation-class.
Not recommended
32GB short
Use this before installing Ollama or LM Studio
This calculator uses conservative rules of thumb for quantized local models. Actual speed depends on model architecture, context length, cooling, drivers, and the app you use to run inference.
AI cost desk
AI Model Pricing Sheet
A worksheet for comparing AI provider costs, hidden pricing drivers, model fit, and budget assumptions without relying on stale static prices.
Provider cost worksheet plus budget notes. Updated when major pricing changes ship.
Use the calculator