can my pc run ai?
Find out which local AI models your computer can handle — Llama, Qwen, Gemma, DeepSeek and more. No download, no sign-up.
What your browser can see
Browsers round memory and cap it at 8 GB, and they never reveal your GPU's video memory (VRAM). Enter your real numbers below for an accurate answer — on Windows: Task Manager → Performance; on Mac: Apple menu → About This Mac.
| Model size | Needs (approx.) | Your PC | Where it runs |
|---|---|---|---|
| 1B | ~1.6 GB | ❌ unlikely | enter your RAM/VRAM to check |
| 3B | ~2.8 GB | ❌ unlikely | enter your RAM/VRAM to check |
| 8B | ~5.8 GB | ❌ unlikely | enter your RAM/VRAM to check |
| 14B | ~9.4 GB | ❌ unlikely | enter your RAM/VRAM to check |
| 32B | ~20.2 GB | ❌ unlikely | enter your RAM/VRAM to check |
| 70B | ~43 GB | ❌ unlikely | enter your RAM/VRAM to check |
Popular open models and what they need
| Model | Parameters | Q4 memory (approx.) | Your PC |
|---|---|---|---|
| Llama 3.2 1B | 1.2B | ~1.7 GB | ❌ unlikely |
| Llama 3.2 3B | 3.2B | ~2.9 GB | ❌ unlikely |
| Llama 3.1 8B | 8B | ~5.8 GB | ❌ unlikely |
| Llama 3.3 70B | 70B | ~43 GB | ❌ unlikely |
| Qwen3 4B | 4B | ~3.4 GB | ❌ unlikely |
| Qwen3 8B | 8.2B | ~5.9 GB | ❌ unlikely |
| Qwen3 14B | 14.8B | ~9.9 GB | ❌ unlikely |
| Qwen3 32B | 32.8B | ~20.7 GB | ❌ unlikely |
| Qwen3 30B-A3B mixture-of-experts: needs memory for all 30B, but runs about as fast as a 3B model | 30.5B | ~19.3 GB | ❌ unlikely |
| Gemma 3 1B | 1B | ~1.6 GB | ❌ unlikely |
| Gemma 3 4B | 4.3B | ~3.6 GB | ❌ unlikely |
| Gemma 3 12B | 12.2B | ~8.3 GB | ❌ unlikely |
| Gemma 3 27B | 27.4B | ~17.4 GB | ❌ unlikely |
| Mistral 7B | 7.3B | ~5.4 GB | ❌ unlikely |
| Mistral Small 24B | 24B | ~15.4 GB | ❌ unlikely |
| Phi-4-mini 3.8B | 3.8B | ~3.3 GB | ❌ unlikely |
| Phi-4 14B | 14.7B | ~9.8 GB | ❌ unlikely |
| DeepSeek-R1-Distill-Qwen 1.5B | 1.8B | ~2.1 GB | ❌ unlikely |
| DeepSeek-R1-Distill-Qwen 7B | 7.6B | ~5.6 GB | ❌ unlikely |
| DeepSeek-R1-Distill-Llama 8B | 8B | ~5.8 GB | ❌ unlikely |
| DeepSeek-R1-Distill-Qwen 14B | 14.8B | ~9.9 GB | ❌ unlikely |
| DeepSeek-R1-Distill-Qwen 32B | 32.8B | ~20.7 GB | ❌ unlikely |
| DeepSeek-R1-Distill-Llama 70B | 70.6B | ~43.4 GB | ❌ unlikely |
How the estimate works
Most people run local models in 4-bit (Q4) quantised form — the default in tools like Ollama and LM Studio — because it cuts memory use by about 4× compared with full 16-bit weights while keeping most of the quality. A useful rule of thumb is 0.6 GB per billion parameters plus about 1 GB of overhead. Long chats, big documents and larger context windows add more on top, so treat the numbers as a minimum.
| Model size | Approx. memory at Q4 | Typical hardware |
|---|---|---|
| 1B | ~1.6 GB | Almost any laptop or recent phone |
| 3B | ~2.8 GB | 8 GB RAM laptop |
| 7–8B | ~5.8 GB | 16 GB RAM, or an 8 GB graphics card |
| 14B | ~9.4 GB | 12–16 GB graphics card, or a 16–24 GB Mac |
| 32B | ~20.2 GB | 24 GB graphics card, or a 32 GB+ Mac |
| 70B | ~43 GB | Two 24 GB GPUs, or a 64 GB+ Mac |
GPU, CPU or Mac — what changes
- Dedicated GPU (NVIDIA / AMD): what matters is VRAM. If the whole model fits, replies come fast. If it doesn't, tools split it between GPU and system RAM, which works but slows down sharply.
- Apple Silicon (M-series) Macs: memory is shared, so the GPU can use most of your RAM. A 32 GB Mac can run models a typical 8 GB graphics card can't.
- CPU only: any model that fits in RAM will run, but expect a few words per second on 7–8B models. Fine for trying things out, slow for daily use.
Getting started
- Use the check above to pick a model size that fits comfortably.
- Install a free local runner such as Ollama or LM Studio.
- Download a 4-bit version of a model in that size — start small (3B–8B) and move up if it's fast enough.
- Close heavy apps (games, many browser tabs) to free memory before loading bigger models.
Local models are private and free to run, but the biggest cloud assistants are still more capable for many tasks. Compare their plans in our AI pricing guide, or count prompt sizes with the token counter.
Frequently asked questions
How much RAM do I need to run an AI model locally?
A rough rule for 4-bit (Q4) models is about 0.6 GB per billion parameters plus ~1 GB overhead: an 8B model needs ~5.8 GB, a 14B ~9.4 GB, a 70B ~43 GB. Longer conversations (bigger context) need extra.
Do I need a graphics card?
No, but it helps a lot. Models that fit in your GPU's video memory run many times faster. Without a GPU, small models (1B–8B) still run on the CPU — just slowly. Apple Silicon Macs share memory between CPU and GPU, so they run fairly large models well.
Why can't the website see my exact RAM or VRAM?
For privacy, browsers round your memory, cap it at 8 GB and don't reveal video memory at all. That's why you can type in your real numbers.
What is WebGPU?
A browser feature that lets web pages use your graphics card. If it's available, small AI models can run directly in a web page without installing anything.
Which app should I use to run models?
Popular free options include Ollama, LM Studio and llama.cpp. They download quantised (usually 4-bit) versions of open models and run them on your own computer.
Is anything sent to a server?
No. The check runs entirely in your browser.