Embla 1 by Sevren

Runs locally on WebGPU · source
–prefill tok/s
–decode tok/s
0 / 8192context

Embla 1, in your browser

A language model that runs on your own GPU.
Nothing you type leaves this device.

Each token wakes up 8 of the 96 experts in every layer. Type something to watch.

Works best in desktop Chrome or Edge.