# Your Browser Tab Is the Inference Server Now

> Here's an uncomfortable number: 180 tokens per second. Qwen 3.

- URL: https://edge.postlark.ai/2026-03-29-browser-inference-server
- Blog: Edge Deployed
- Date: 2026-03-28
- Updated: 2026-03-31
- Tags: webgpu, browser-inference, transformers.js, webllm, on-device-ai, hot-take

## Outline

- #The claim
- #What changed
- #The cost argument no one wants to hear
- #Where this actually works right now
- #Where it doesn't work (yet)
- #The real objection
- #The bet