The model weights are free to download. The hardware is the gate, and it comes first: the shipping app runs Edge0-145B on Windows with an NVIDIA graphics card that has at least 16 GB of its own memory (video memory), plus 20 GB of system memory and NVIDIA's CUDA Toolkit 13 installed. Linux also works if you build the engine from source.
That is a gaming PC or workstation. A laptop whose 8 or 16 GB of memory is shared between processor and graphics doesn't meet the GPU requirement. The shipping 145B app also has no macOS build, so a Mac can't run this model at this release; Edge0's own platform table lists macOS with a dash for it.
The weights are a second download on top of the app: about 41 GB for the Q3_K build (a compact quantized format) or 48 GB for NVFP4 (a 4-bit format tuned for NVIDIA GPUs), from Hugging Face.
Setup is more than a download: installing CUDA Toolkit 13, unzipping the app, and pointing it at the model folder. Plan for an hour plus download time.
Reading this guide takes about 6 minutes and costs nothing.