I tried out the Built-in AI that apparently works on Chrome Dev or Canary v127+.
https://developer.chrome.com/docs/ai/built-in

Enable the flag

chrome://flags/
Prompt API for Gemini Nano: enable
Enables optimization guide on device: enable BypassPrefRequirements

Download the model

chrome://components/
After enabling the above flags and restarting the browser, you can access the components and download the model.

The model download itself takes a bit of time. (It took about 1-2 minutes for me.)

Try it from the console.

jsCopywindow.ai
const ts = ai.createTextSession()
const gemi = await ts
const output = gemi.prompt('can you tell me the capital of Japan?')

output

also, it worked as expected in Japanese