EN
7 November 2026 · Jakub M.
New: a RAM calculator for local AI models
Which AI models will run on your Mac, and how much memory does the model you want need? Our new calculator shows it at a glance for 63 open models, from tiny ones to giants with trillions of parameters.
We have launched a RAM calculator. It shows at a glance which open AI models you can run on which Mac, so you don’t have to work it out by hand.
Two ways to use it
- My Mac has: pick how much memory your Mac has, from 16 to 512 GB, and see every model that fits, largest first.
- I want to run: pick a model and see how much memory it needs and which Macs it will run on.
What fits where
| Mac memory | The largest models that fit (4-bit) |
|---|---|
| 16 GB | Ministral 3 14B, Qwen3 8B, Llama 3.1 8B |
| 32 GB | Gemma 4 31B, Qwen3 30B-A3B, gpt-oss-20b |
| 64 GB | Qwen2.5 72B, Llama 3.3 70B, Qwen3.5 35B-A3B |
| 128 GB | Mistral Large 2, Qwen3.5 122B-A10B, gpt-oss-120b |
| 256 GB | DeepSeek V4 Flash, Qwen3 235B-A22B, MiniMax M2 |
| 512 GB | Nemotron 3 Ultra, Qwen3-Coder 480B-A35B, Llama 3.1 405B |
Estimate from the calculator: 4-bit quantization, context up to about 32,000 tokens, default macOS limit for the GPU.
How it calculates
The calculator covers 63 open models, from Qwen3 0.6B to Qwen3.8 with 2.4 trillion parameters. It assumes 4-bit quantization, the usual choice for local models, and adds a reserve for the context and the runtime. It also takes into account that by default macOS gives the graphics only about two thirds to three quarters of the memory. The results are estimates: actual use depends on the software and the length of the context. The maths is explained in our article on how much memory a language model needs.
Try it
Open the calculator. If your own Mac isn’t enough, you can rent a Mac Studio M5 Ultra with 256 GB from one day.