Tutorials·beginner·8 min

Voice calls in MultiChats: which AI model is actually on the line

How MultiChats voice calls work on your phone: Fast Voice or Any Model, which models take a call, what the chat keeps and each plan's limits.

Multi Chats Team
November 13, 2026

Saturday, a quarter to nine, and Inès is on the towpath with the dog lead in one hand and her phone in the other. Her dad turns seventy in three weeks. On Thursday she typed the guest list, two restaurant names and a budget into a MultiChats chat with Claude Opus 5.5, then ran out of evening. Now she has forty minutes of walking and no free hand for typing, so she opens that chat, taps the small sound-wave icon beside the message box and starts talking.

It had been a loud week for voice. On 14 October xAI announced Grok Voice 3.0, a speech-to-speech model, and on 15 October OpenAI began rolling out GPT-Live-1.1 in ChatGPT Voice (both checked on 19 October 2026). In both cases the voice and the model behind it come from one company. MultiChats makes a bigger claim with its "Any Model" mode, so we checked who picks up.

The name at the top of the call screen

The first clue sits above the animated orb: "Gemini Flash", with a small lightning bolt. That is Fast Voice, the default mode, subtitled "Lowest latency". It runs on Google's Gemini Flash native-audio model, which hears and speaks without handing off between a transcriber, a writer and a reader. That is where the speed comes from. The model Inès picked for her chat plays no part in the call: on Fast Voice, every call goes to the same Gemini model.

Ask the voice what it is and it will not say Gemini: our instructions tell it to call itself the MultiChats assistant. The label is the reliable answer.

Fast Voice can also use Google Search during the call. When Inès asks whether the second restaurant serves Sunday lunch, it can look it up. A spoken summary of a search is still a summary; she will check the booking page later.

To start a call, open any chat in the iPhone or Android app and leave the message box empty. The sound-wave icon sits next to the microphone; once you type, both give way to the send arrow.

Want to follow along as you read? Open MultiChats

What "Any Model" connects you to

The other mode, Any Model, is subtitled "Full flexibility" and is on Pro and Pro+ only. Deepgram turns your speech into text, the model you picked writes the reply, and a separate synthetic voice reads it out. The screen shows the model's name above "via Deepgram · Cartesia Sonic 3". Any Model is slower than Fast Voice because it has three stages.

"Any" is generous. The mode works with models that support voice calls, and as of 19 October 2026 that is a short list, mostly from the previous generation: GPT-5.4 and GPT-5.4 Mini, Claude Sonnet 4.6 and 4.5, Gemini 3 Flash, Gemini 3.1 Flash Lite and Gemini 2.5 Flash Lite. The newest flagships are not on it. When Inès, who is on Pro, tries Any Model from her Claude Opus 5.5 chat, the app says "Claude Opus 5.5 doesn't support voice mode yet." and offers "Switch to Claude Sonnet 4.6". From a GPT-6 model, or from providers such as DeepSeek or Qwen, the offer is GPT-5.4. The switch applies to the call only; the chat keeps its model for typing.

If the chosen model fails to answer mid-call, a backup steps in and an amber line beginning "via" names it.

Talking, interrupting and the silence rule

The line under the orb shows the call's status: "Connecting...", "Listening...", and "Thinking..." while the model works something out. While it speaks the line reads "Tap to interrupt", and a tap on the orb stops it mid-sentence. There is a mute button, and the red button hangs up. The call screen has no live captions at the moment; the words turn up in the chat instead.

Go quiet for about 45 seconds and the screen asks "Still there? Speak or tap to continue". Fifteen seconds later, the call ends.

We tell the voice to keep answers to two or three spoken sentences unless you ask for more. If something only works on screen, such as a table, it gives a short spoken version and suggests the text chat. Ask for the full menu comparison and that is what happens, which is the voice admitting that a table read aloud helps nobody.

The call needs the app open on screen. Switch to another app or lock the phone and MultiChats hangs up. Everything said so far is saved. Before a long walk, check how quickly your phone locks itself.

Voice Settings, and one description worth a second look

The gear icon opens Voice Settings. "Voice Mode" switches between Fast Voice and Any Model; a free account gets Fast Voice only. "Voice" lists presets for the current mode, such as Warm or Gentle on Fast Voice and British Lady or Calm Lady on Any Model.

"Pace" offers Slow, Normal and Fast, with Fast as the default. On Fast Voice it sets how long the model waits after you stop talking, so Slow leaves room to pause mid-thought. The voice itself speaks at the same speed.

Two settings appear on Fast Voice only. "Thinking" sets how much the model reasons before it speaks, and Auto, the default, suits a walk. "Proactive Audio" is the interesting one. Our app describes it as "Model responds naturally without waiting for silence". Google's documentation for the feature puts it differently:

"When this feature is enabled, Gemini can proactively decide not to respond if the content is not relevant."

On a towpath with cyclists ringing bells, that is the useful reading: with it on, the model may let a passing "morning!" go unanswered. Changing either setting during a call restarts the call, with a brief "Restarting call with new settings" notice.

The speaker icon opens "Audio Output": "Speaker" (the default) or "Phone", the earpiece. The other settings also sit under Voice in your profile.

After the walk, the chat keeps the words

Each turn of the call, hers and the assistant's, is saved to the same chat as an ordinary text message as soon as it finishes. In the phone app the replies are labelled with the model that spoke them, "Gemini Flash" for a Fast Voice call, followed by a small sound-wave mark. The chat saves the words, but not the audio.

It works the other way too. A call opens with the recent part of the chat as background, which is how the voice knew Thursday's guest list. Attached files reach the call only by name, so the voice cannot open the menu PDF itself, though it knows what was said about it.

When you type, the chat uses your chosen model. Back home, Inès types to Claude Opus 5.5: "Turn what we agreed on the walk into an invitation for the family group." Opus reads the transcript like any other part of the conversation. The call ran on Gemini; she trusts Opus with the writing.

The microphone beside the sound-wave icon is dictation: your words land in the message box, ready to tidy and send to whichever model the chat uses. Dictation is on the website too. Calls are in the iPhone and Android apps only.

The limits, as of 19 October 2026

A free account gets Fast Voice only, calls of up to 3 minutes and 10 minutes of calls per calendar month, plus dictation clips of up to 2 minutes. Pro gets both modes, calls of up to 30 minutes and 120 minutes a month, and dictation clips of up to 5 minutes. Pro+ has the same call limits as Pro, with dictation clips of up to 10 minutes. Call time is counted in minutes, separately from your messages. On the free plan a call sets aside its full three minutes when it starts, and any unused time comes back within a few minutes.

The app shows no running count. At a limit the call ends and a line appears above the message box for a few seconds: "Duration limit reached." or "Monthly voice limit reached." It then says what to do next, such as "Your minutes reset next month." If the app says "Please update the app to use voice calls.", updating is the fix.

Our verdict

Use Fast Voice on a walk. It is quick and can search, and the screen names Gemini. Any Model earns its place when you want a particular model in your ear, if it is on the voice list and you are on Pro or Pro+. The label promises more than the list delivers, and we would rather you heard that from us.

The transcript is what matters. A MultiChats call is a first draft spoken out loud, and whichever model you pick for the chat can take it from there. The October launches compete on sounding human. With a dog and a birthday to plan, knowing who was on the line and keeping every word proved more useful. By the last lamp post, Inès had picked the restaurant, and the invitation was waiting in the chat to be written.