Talk to realtime voice models from OpenAI, Google, xAI and ElevenLabs in Phaseo Chat

A voice model's benchmark scores don't tell you how long it takes to answer, whether it stops when you interrupt it, or what ten minutes of conversation cost. You find out by talking to it.

Every Phaseo account can now make realtime voice calls in Chat → Realtime. Pick a model and a voice, allow microphone access, and start talking. You don't need to install anything or create an API key. Phaseo charges each call to your workspace credits.

Voice models from four providers

To compare two models, ask each one the same question with the same instructions, one call after the other.

The cost of each call, while it runs

Chat shows how long the call has run and how much it has cost so far. When the call ends, Chat shows the final bill. Prices are the model's published prices, the same as on the API.

Chat ends a call after five minutes with no activity, so a tab you forgot about stops charging you. A call also ends when your key reaches a spend limit. Each provider sets a maximum call length, from 25 minutes for GPT Live to two hours for Grok Voice.

What you can change

Each model offers its own built-in voices. For GPT Live voices, Chat also shows the accent and style the provider documents.

You can give the model instructions, such as who it is and how short its answers should be. Chat keeps them when you switch models.

For GPT Live, you can choose the background model, its reasoning effort and output limit, and whether it can search the web.

Realtime voice on the API

Phaseo is opening realtime voice on the API to a small number of workspaces first. The realtime voice guide explains how sessions, events and billing work there.

Get started

Open Chat → Realtime. If your workspace has no credits yet, add some in Settings → Credits first.