Guide · GAOBA AI Resources
AI Models That Fit Your RAM
A short guide (in Arabic) to choosing an AI model you can run on your own computer by RAM size: 8, 16, 32 or 64 GB, with each model's size, its Ollama run command and its link.
If the PDF does not open inside Instagram, tap ⋯ (top corner) and choose “Open in browser”.
The golden rule: pick a model clearly smaller than your RAM, because the operating system and open apps need part of it too.
You can run an AI model on your laptop with no internet and no subscription, but the model has to fit your RAM. This guide sorts models by RAM: 8, 16, 32 and 64 GB.
For each model: its size, its Ollama run command and its official page. Before that, 3 rules to help you choose well, and two ways to run models: a command line or an easy app.
What you get
- Know which model fits your computer: 8, 16, 32 or 64 GB.
- 11 models, each with its size, run command and page.
- Two ways to run them: Ollama on the command line, or LM Studio with an easy interface.
- 3 rules before downloading, so your computer doesn't slow to a crawl.
- How to check your RAM on Windows and Mac.
- A clear note on models too big for a laptop, such as Kimi K3.
How much RAM do you have?
Before you choose, check how much RAM your computer has:
The rules 3
Rule 1
The listed size = the compressed (4-bit) download in Ollama
The model needs about the same amount of memory, and a bit more in long conversations.
Rule 2
Leave room for your computer
The operating system and open apps take part of the RAM, so pick a model clearly smaller than your RAM.
Rule 3
A graphics card (GPU) speeds things up
Without a GPU the model runs on the processor, just slower. On Macs with Apple chips, the RAM is shared with the graphics.
Two ways to run them
Pick one: a command line or an easy app. Both run on Windows and Mac.
Ollama
- Command line: install it, then type the command listed next to the model, e.g. ollama run qwen3:4b.
- Every run command on this page is for Ollama and has a copy button.
LM Studio
- Easy interface: install it and search for the model's name inside the app.
Models by RAM
8 GB RAM: small models
Qwen3 4B
Size: 2.5 GBModel pageollama run qwen3:4bGemma 3 4B
Size: 3.4 GBModel pageollama run gemma3:4b
- Close heavy apps while it runs. Models over 5 GB are heavy on 8 GB.
16 GB RAM: the popular models
Llama 3.1 8B
Size: 4.9 GBModel pageollama run llama3.1:8bQwen3 8B
Size: 5.2 GBModel pageollama run qwen3:8bGemma 3 12B
Size: 8.2 GBModel pageollama run gemma3:12bQwen3 14B
Size: 9.3 GBModel pageollama run qwen3:14bGPT-OSS 20B
Size: 14 GBModel pageollama run gpt-oss:20b
- GPT-OSS 20B on 16 GB is tight: it runs, but leaves your computer little room.
32 GB RAM: large models
Gemma 3 27B
Size: 17 GBModel pageollama run gemma3:27bQwen3 30B
Size: 19 GBModel pageollama run qwen3:30bQwen3 32B
Size: 20 GBModel pageollama run qwen3:32b
64 GB RAM: the heavyweights
Llama 3.3 70B
Size: 43 GBModel pageollama run llama3.3:70b
- Models this size take patience: speed depends a lot on the graphics card.
- GPT-OSS 120B is 65 GB, more than 64 GB of RAM, so it doesn't fit this machine.
What about Kimi K3?
One of its compressed versions (Unsloth 1-bit) needs about 610 GB of memory (RAM + VRAM combined), so it's not for a laptop.
Source: UnslothWhere the sizes come from
The Ollama library (each model's tags page), reviewed on 9 November 2026. Sizes change with updates, so check the model's page before downloading.
What you need
- A Windows or Mac computer; the more RAM, the more models you can run.
- Ollama or LM Studio.
- Enough storage for the model you download.
Keep the guide with you
Everything on this page, in one PDF to keep on your phone or computer.