Grok Voice: spoken conversations

Updated November 3, 2026, based on the xAI Text to Speech, Speech to Speech, Pricing, and Rate limits pages: 28 built-in voices, current prices, concurrent sessions by tier, and a code example that uses the xAI POST /v1/tts endpoint. Sources: Text to Speech · Pricing · Rate limits.

Talk with Grok

After text and images comes voice. You can hold real spoken conversations with the assistant, and right away you need to tell apart two services that people often confuse: the Voice Agent, designed for interactive real-time conversations, and Text-to-Speech (TTS), which simply turns text into audio. The first listens and answers; the second reads. Picking the wrong one is the most common reason voice projects disappoint.

Voice Agent: real-time conversation

The Voice Agent is the most immersive service: it lets you talk directly with Grok as you would with a person. It runs on a WebSocket connection that carries two-way audio in real time. You speak into your microphone, Grok listens and understands your message continuously, then answers out loud; turn detection, or VAD, handles silences and transitions automatically. That last point is what separates a conversation from an exchange of voice messages: you can hesitate without the system thinking you’re done, and interrupt it when it heads in the wrong direction.

Twenty-eight built-in voices are available (as of November 3, 2026), each with its own personality, and you should choose based on the use case rather than personal taste. The five original voices, as xAI describes them:

The other twenty-three, more recent, are grouped by xAI by use case (support, narration, education, advertising, wellness) on the Text to Speech page of its documentation.

Delivery goes beyond the timbre of the voice. Expressive tags enrich the answers: pauses set a natural rhythm, laughter lightens a moment, whispering creates a sense of closeness. Used sparingly, they make the difference between a passable voice and one people listen to until the end; in every sentence, they have the opposite effect.

The Voice Agent can also call tools during the conversation: real-time web searches, database lookups, calls to custom APIs. You ask out loud for up-to-date information, and Grok answers out loud with the result. A support assistant can check the actual status of an order instead of reciting a generic procedure.

Text-to-Speech: reading text aloud

TTS meets a simpler need: turning written text into spoken audio. It’s used to generate narration for videos or podcasts, produce automated voice messages, make text content accessible to people with low vision, and broadcast voice announcements and notifications. The service covers twenty languages and can also detect the language automatically.

The API call takes just a few lines:

import os
import requests

response = requests.post(
    "https://api.x.ai/v1/tts",
    headers={"Authorization": f"Bearer {os.environ['XAI_API_KEY']}"},
    json={
        "text": "Welcome to this course on Grok.",
        "voice_id": "eve",
        "language": "en",
    },
)
with open("welcome.mp3", "wb") as f:
    f.write(response.content)

The result is an MP3 audio file, ready to play or to drop into an edit.

What voice changes for businesses

The first area is customer service: deployed as a voice assistant, the Voice Agent handles common questions and runs real-time searches, absorbing repetitive volume without making callers wait. The same logic extends to existing telephony: xAI can answer your numbers itself over SIP, and frameworks such as LiveKit or Pipecat plug in, without rebuilding your call infrastructure.

TTS mainly serves content production: your articles, reports, and presentations get an audio version, with a voice chosen for the tone you want. In both cases, accessibility deserves to be part of the design from the start rather than added afterward: these capabilities make AI usable for people that text interfaces leave out.

What it costs

The Voice Agent costs $0.08 per minute, or $4.80 per hour, plus $0.004 per text input; on the entry tier, a team can open 10 concurrent sessions, and up to 200 on higher tiers. Text-to-Speech costs $15 per million characters, with 50 requests per second and 100 concurrent sessions on the entry tier. Check these concurrency caps before any public launch: they determine how many calls you can handle in parallel, and a campaign that exceeds them results in callers on hold, not a higher bill.

Prices checked November 3, 2026. Prices change often: before you budget anything, check the current rates on the official xAI pricing page.

Key takeaways

• The Voice Agent enables real-time spoken conversations with Grok over WebSocket
• 28 built-in voices are available, with different personalities and use cases
• TTS converts text to audio (MP3 by default) in 20 languages
• The Voice Agent can use tools (web search, APIs) during the conversation
• Standard telephony protocols are supported for business integration