Browser GPU Benchmark

WebGPU

Runs SmolLM2 360M entirely in your browser using WebGPU. Measures real decode tok/s — the number that determines how fast you can actually chat with AI. No server, no tracking.

ModelSmolLM2-360M-Instruct (q4f16)

Download size~376 MB (cached after first run)

What is measuredDecode tok/s for 128 output tokens

RequiresChrome 113+ or Edge 113+ with WebGPU enabled

Why SmolLM2 360M? It’s the smallest model with a strong quality/speed signal — fast to download, runs everywhere WebGPU works, and its tok/s is a reliable indicator of what larger models will do on the same GPU.

Privacy: The model runs 100% in your browser. No prompts or outputs leave your device. Submitted benchmark numbers are GPU + tok/s only, no PII.

Powered by: WebLLM (MLC AI, MIT) + MLC-LLM

Want to check what AI models your hardware can run? Auto-detect on the homepage →