Can I run AI on 8GB RAM?

Yes, with small models. An 8 GB laptop can run AI privately for quick questions, summaries and simple writing. Just don't expect ChatGPT-level answers.

Partly tested Oct 2026 (one small model, CPU only). We got part of the way in on the free plan before a paid plan or trial was required; see “What happened when we tried it”. Everything else comes from official pages (last checked Oct 2026). How we test

Short answer

Yes. Pick a model whose download is a few GB smaller than your memory. On 8 GB, these fit (Ollama download sizes):

Llama 3.2 3B: 2.0 GB
Qwen 3.5 4B: 3.3 GB
Gemma 4 E2B: 4.6 GB (tighter)

They're good for quick questions, summaries and simple writing.

What to expect

Simpler answers than big cloud chatbots, and slower replies on modest hardware.
Close other heavy apps (browsers with lots of tabs) to leave memory spare.
Long documents need extra memory, so keep inputs short.

Which app?

Mac (M1 or newer): LM Studio says 8 GB Macs work with smaller models (it recommends 16 GB+). Ollama works too.
Windows: LM Studio recommends 16 GB, so on 8 GB use the smallest models; Ollama is also an option.
GPT4All says 8 GB is enough for 3B models.

What we actually ran: Qwen 3.5 0.8B with Ollama on a CPU-only Linux server. We haven't tested the 3–4B models on an 8 GB laptop ourselves.

Check your exact computer →

Next: Full local AI guide · Ollama vs LM Studio

Sources

Ollama: Qwen 3.5 sizes: last checked Oct 15, 2026
Ollama: Gemma 4 sizes: last checked Oct 15, 2026
Ollama: Llama 3.2 sizes: last checked Oct 15, 2026
LM Studio: system requirements: last checked Oct 15, 2026
GPT4All: system requirements (GitHub): last checked Oct 15, 2026

Prices and plans change often. Always confirm on the official site before you buy.