Voice calls now run on GPT-Live

Voice calls in Bearly now run on GPT-Live, OpenAI's full-duplex voice model, on paid plans. It listens while it speaks, so you can cut in without waiting for a pause. When a request needs a tool, such as a web search, Bearly runs it in the chat while the conversation continues. Since this post went out, paid plans can also switch calls to Gemini 3.8 Live or Gemini 3.8 Live Extended Thinking from Settings; details are below.

Calls start the same way as before: open a chat and select Start voice call, the microphone button beside the message box.

It hears you while it talks

Bearly keeps listening while it answers. If it starts explaining something you already know, say so. If you remember a detail halfway through its reply, add it. You don't have to wait for it to finish.

Because it is always listening, nearby voices can start a reply you didn't ask for. Headphones help, and Mute microphone stops sending your audio until you're ready to speak again.

The call keeps going while Bearly works

Say you're preparing for a customer meeting, and the chat already holds your notes from the last one. Start a call and talk through what you want to cover. The call starts with the recent messages in that chat, so Bearly can pick up where your notes left off.

Partway through, ask Bearly to check whether the customer has announced anything new this month. Bearly runs the web search in the chat. While it runs, you can keep going through the agenda, and Bearly answers when the results come back. The search appears in the conversation as it happens and stays there after the call.

Web search is one example. A call can use the tools available in that chat.

What the chat keeps

When you end the call, it stays in the chat as a card with its length and a Show transcript control. Bearly saves the transcript, not a recording of your voice.

The transcript becomes context for the rest of the chat. Follow the call with a written request, such as "Turn what we discussed into a one-page brief with open questions," and Bearly works from what was said.

Pick a different voice model

GPT-Live 1 is the default. On a paid plan, open App Settings → Chats → Voice and choose Gemini 3.8 Live or Gemini 3.8 Live Extended Thinking instead. The change saves as soon as you pick it, follows you to your other devices, and applies to your next call; a call already in progress keeps its model.

Both Gemini models listen while they speak and use the chat's tools the same way. Extended Thinking reasons more before it answers, which suits questions with several steps. Two things differ on Gemini calls: permission requests appear as cards in the chat rather than as spoken questions, and the voice you hear is the model's own, not the GPT-Live voice.

Where it works

Voice works in the Bearly desktop app and in the web app, except in Firefox, where it is temporarily unavailable. It needs microphone access. On a phone, keep Bearly in the foreground during a call; the operating system can pause the microphone and network when the app is in the background.

Paid plans run calls on GPT-Live 1 unless you choose a Gemini model in Settings; Free accounts keep a smaller voice model and cannot switch. Each call has a usage allowance that depends on your plan. When a call reaches it, Bearly shows Voice limit reached and keeps the transcript so far. A call also ends after 30 minutes, or sooner if it stays quiet. Start another call to keep going.

On a managed team, an administrator can turn voice calls off.

The voice conversations guide covers the call controls and troubleshooting.