Talk to Gemma 4

checking…

Record a voice memo, drop in an image, or just type. Audio goes straight into the model — no transcription step in front, no separate vision endpoint beside it. Served by Plubo on an OpenAI-compatible API. No key needed here.

Say something, drop an image, or hold a thought out loud.

Audio is capped at 30 seconds — that is the model's limit, not the demo's. Gemma 4's audio ability is speech recognition and speech translation, per Google's model card; it is not a tone or emotion classifier.

ModelContextInput$/1M in · outImage · audio $/1M
gemma-4-12b-it262,144text · image · audio · video
gemma-4-e4b-it131,072text · image · audio · video
gemma-4-e2b-it131,072text · image · audio · video