Talk with an open model
Speak, and the answer comes back in your ear. Speech recognition (Qwen3-ASR), a small open language model (Qwen3 4B) and speech synthesis (Kokoro) all run on Babel Mesh nodes.
You need a trial key first. Claim free credits on the chat page, then come back here: the same key works in this browser.
Press Start talking, then speak. The demo answers when you stop.
Your voice is sent to the speech engine for each turn, and the text of the conversation is sent to the language model. Nothing is stored on the server; the conversation lives in this page until you leave it. Use a headset or turn the volume down so the answer does not reach the microphone.