An 8 GB computer can test local AI, but it needs a small quantized model, short prompts, and enough free memory for the operating system.
What should you choose?
For most 8 GB computers, Qwen 3 4B in a Q4 format is the best balanced starting point. Choose Qwen 2.5 Coder 3B for coding, Phi-4 Mini for compact reasoning, or Qwen 3 1.7B when the computer struggles with larger models.
This answer assumes the computer has 8 GB of total RAM, no large apps open, one user, a short conversation, and a Q4-size model file.
Qwen 3 4B Q4
It offers a useful balance of writing, reasoning, coding, and several languages while remaining small enough for a careful 8 GB test.
Limit: It will be tight on Windows. Close other apps and keep the context short.ollama run qwen3:4bQwen 2.5 Coder 3B Q4
It is smaller than many coding models and is focused on code completion, explanation, and simple fixes.
Limit: It is not the right size for large repositories or difficult multi-file work.ollama run qwen2.5-coder:3bPhi-4 Mini 3.8B Q4
It is a small model with useful reasoning and math ability for its size.
Limit: Long prompts and difficult research tasks can still overwhelm it.ollama run phi4-miniQwen 3 1.7B
It leaves more room for Windows, a browser, and the AI engine.
Limit: Answer quality is lower than the 4B choices on difficult work.ollama run qwen3:1.7bDo not begin with 7B, 8B, or 14B models on an 8 GB computer. A file may download successfully and still leave too little working memory.
8GB RAM is Enough: The Truth About Local AI
This Deviceterra video shows why the exact model file, free system memory, and real task matter more than the 8 GB label alone.
Use the explanations below when you want to know why each step matters.
What 8 GB really means
Your AI model cannot use all 8 GB. Windows, macOS, or Linux already uses part of the memory. Your browser and other programs use more. This is why a model file that looks small enough can still make the computer slow.
Treat 8 GB as an entry-level test machine. It can help with short summaries, rewriting, simple questions, and light coding help. It is not a good starting point for many users, large documents, or complex agents.
Start between 1B and 4B
Begin with an instruction model between about 1 billion and 4 billion parameters. A Q4 quantized file is often the safest first choice because it uses less memory than a full-precision model.
Small Gemma, Qwen, Phi, and Llama family models may fit this range. The exact file matters more than the family name. Check the file size and quantization before downloading.
Start small. Move up only when the smaller model cannot complete your real task.
Prepare the computer
- Close games, video editors, and browser tabs you do not need.
- Keep several gigabytes of free storage beyond the model file.
- Start with a short conversation and one document at a time.
- Use a CPU-friendly engine such as Ollama, LM Studio, or llama.cpp.
- Restart the AI app if memory use keeps growing after a long session.
Run a useful test
Create five prompts from work you really do. Ask for a short summary, a rewrite, a list, a simple explanation, and one task-specific answer. Record whether the answer is correct and how long you wait.
Do not judge the model from one impressive answer. If it fails the same type of task more than once, try a better prompt or a different small model.
When 8 GB is not enough
Move to a 16 GB or larger computer when you need longer documents, stronger coding help, image understanding, several apps open at once, or faster answers.
Use the MamiLens hardware path to see conservative matches. If no model has enough memory headroom, MamiLens should say so instead of forcing a recommendation.
Try a safe model on an 8 GB computer
An 8 GB computer can run local AI, but the operating system already uses part of that memory. Start small and leave space for Windows, macOS, or Linux.
- 1
Check total memory
On Windows, press Ctrl, Shift, and Esc. Select Performance, then Memory. Confirm that the total is about 8 GB.
- 2
Close heavy programs
Save your work. Close games, editing apps, and browser tabs you do not need.
- 3
Install Ollama
Visit ollama.com/download, choose your system, and run the installer.
- 4
Open a command window
Windows: click Start, type PowerShell, and open it. Mac: press Command and Space, type Terminal, and press Enter.
- 5
Run a 1B model
Type this command and press Enter.
ollama run gemma3:1b - 6
Ask one short question
Ask for a five-sentence summary. Avoid long documents during the first test.
- The computer remains responsive.
- The model completes a short task.
- You still have free memory while it runs.
Official facts and real user evidence
Official documentation supports product and model facts. Community discussions show real setups, failures, and questions. A community result is supporting evidence, not a promise that another computer will perform the same way.
Find a model your computer can run.
MamiLens checks your hardware and shows a careful starting point.
Run the free compatibility check →This guide is educational. Model software, licenses, and hardware support can change. Check official sources before an important deployment.