Open-weights translation running on our own hardware
The translation will appear here.
Real-world snippets, already under the character limit.
Loading examples…
Translation and speech transcription run locally behind this API. The optional read-aloud control uses your browser or operating system voice, which may depend on a platform speech service.
Every response reports how long it took, how many tokens were generated, and at what rate. This is a CPU deployment — a GPU makes it roughly an order of magnitude faster.
The dropdown lists only the languages the model was formally benchmarked on (WMT24++). It accepts more, but those are the ones with published quality evidence.
Powered by TranslateGemma, an open-weights model from Google, used under the Gemma Terms of Use and subject to Google's Prohibited Use Policy.