diciz runs a real language model inside this browser tab, on your own GPU through WebGPU. Paste the URL of a GGUF model or pick a preset: the weights download once and are cached in your browser, then the chat works offline. Your messages stay in the tab. Measured on 2026-06-10 on an NVIDIA RTX 4070 Laptop GPU (8 GB) in Chrome: Qwen2.5-7B (Q4_K_M) at 31.2 tok/s, Qwen2.5-0.5B at 94.2 tok/s (method and raw results). Also in diciz: video frame generation. Made by ZMLabs · diciz.com · Privacy.