AI

Which AI model can my computer run?

Tell the finder what computer you have and it lists the open-weight AI models that will run on it, best first. Each model comes with the quality level that fits your memory, a speed estimate and a command to download it. The list is refreshed every day from Hugging Face, so recent releases are included.

Your choices stay on this page. Only the model list is downloaded, the same for every visitor.

Makers
All makers
Only show

Loading the model list.

    How to use it

    1. Pick your computer: a Mac with Apple silicon, a PC with a graphics card, or a computer without one.
    2. Choose how much memory it has. On a Mac, that is the unified memory. On a PC, choose the graphics card memory (VRAM) and the system memory (RAM).
    3. Pick what you want the model for, or leave it on Chat and writing.
    4. Read the list from the top. Recommended order puts capable, widely used models first, each at the best quality that still runs at a good speed. Sort by Best quality, Fastest, Newest or Most popular to change that.
    5. Narrow the list if you like: only Good speed or faster, open licenses, chosen makers, MoE models, models that see images, or releases from the last 30 days. The address keeps your settings, so you can bookmark or share the view.
    6. Open a model on Hugging Face, or copy its Ollama command and paste it into a terminal to download and start it.

    Questions

    How do I find out how much memory my computer has?

    On a Mac, open the Apple menu and choose About This Mac; Memory is the unified memory. On Windows, open Task Manager, go to Performance and select GPU: Dedicated GPU memory is your VRAM, and the Memory page shows your RAM. On Linux with an NVIDIA card, run nvidia-smi.

    What do Q4, Q8 and BF16 mean?

    They are quality levels, also called quantization. BF16 is the model as released, at 2 bytes per parameter. Q8 squeezes it to about half that size with almost no loss. Q4 brings it under a third of the size with a small loss that most people do not notice in chat. By default the finder picks the highest level that still runs at a good speed on your computer, and notes when a higher one also fits. Sort by Best quality to always see the highest level that fits.

    Why does a Mac only get 70% of its memory?

    A Mac shares one pool of memory between the system, your apps and the graphics chip. By default macOS lets the graphics chip use roughly two thirds to three quarters of it, and you still need room for a browser and the app that runs the model. The finder plans with 70%.

    What does MoE mean, and why is it faster?

    MoE stands for mixture of experts. The model is split into many small expert networks and uses only a few of them for each word, so a 35B MoE model that uses 3B parameters at a time runs about as fast as a 3B model. It still needs memory for all of its parameters.

    My graphics card has 8 GB. Why do bigger models still show up?

    Apps such as Ollama and LM Studio can keep part of a model in system RAM when it does not fit in the graphics card. That works, but system RAM is many times slower than graphics memory, so those models are marked Part in system RAM and listed after the ones that fit on the card.

    How is the speed worked out?

    Every word a model writes means reading its active weights from memory once, so speed depends mostly on memory speed. The finder uses typical memory speeds for each kind of computer. Fast means the text appears quicker than you can read it, Good speed is comfortable for chat, Slow is usable if you can wait, and Very slow means a few words a second or less. Your chip, the app and long conversations all change the real figure.

    What counts as an open license?

    Apache 2.0, MIT, BSD and similar licenses let you use, change and sell what you build with few conditions. Some labs publish their own terms instead, such as the Llama and Gemma licenses, which add rules on use or on very large companies. Those are marked Custom license, and non-commercial ones are marked as such. Read the license on the model's page before you use a model in a product.

    Is anything about my computer sent anywhere?

    No. The page downloads the same model list for every visitor and works out the fit in your browser. The memory and use you pick stay on this page, and your last choice is remembered only in this browser.

    Where does the model list come from?

    It holds the most downloaded open-weight models from the major AI labs on Hugging Face, such as Qwen, Google, DeepSeek, NVIDIA and OpenAI, and is refreshed every morning. Memory figures are estimates for the model weights plus about 1.5 GB for the conversation. Very long documents or chats need more.

    Advertisement