What is on your mind?
27 billion parameters at ternary precision, decoded by WebGPU kernels on your own GPU. Nothing you type leaves this machine.
Hold to start
27B ternary weights · 5.9 GB download, cached after the first visit
~9.3× smaller than FP16 · 98.2% of FP16 intelligence retained
Loading model
0%
Private preview
Paste a Hugging Face access token with read access to the model repo. It stays in this browser — requests go only to huggingface.co.
27 billion parameters at ternary precision, decoded by WebGPU kernels on your own GPU. Nothing you type leaves this machine.