Voice Mode, End to End

Voice mode is a real-time conversation with any model in the catalog: you speak, it answers aloud, and you can talk over it to cut it off like a person. Here is how the pieces fit and where voice genuinely beats typing.

The pieces

• Speech-to-text: your words stream to the model as you say them (transcription also works standalone, for recordings and dictation).
• The model: voice is model-independent: pick the brain separately from the voice. A fast model makes conversation feel instant; see the cheap models roundup for good low-latency picks.
• Text-to-speech: replies come back in your chosen voice from the catalog, searchable by language and style.
• Barge-in: interrupt mid-reply; the model stops and listens. This one behavior is what makes it a conversation instead of a dictaphone.

Everything lands in the chat as text too: a voice session is a normal conversation you can read, search, and continue by keyboard later.

Where voice wins

• Thinking out loud: brainstorming and untangling problems, where typing speed throttles thought.
• Hands busy: cooking, driving, whiteboarding: ask and hear back.
• Drafting by dictation: capture a rough draft by voice, then have an editing model tighten it in the same thread.
• Two-second questions: on Windows, the Voice HUD makes these a global hotkey, no window switch.

Good to know

• The microphone is open only during an active voice session.
• Any text reply in any chat can be played aloud with the speaker control, in your chosen voice.
• Voice, model, and playback settings live per chat; the help article covers each control.

Try one full voice session on a problem you would normally type at: the surprise is not the transcription accuracy, it is how differently you think when you talk. The voice mode feature page has the full picture.