A Grok Voice 3.0 agent costs $0.08 per minute of session, so a 5-minute call is $0.40 and 1,000 minutes is $80. On top of that, xAI charges $0.004 each time your code sends a text item into the live session. That is the whole price list for the model itself; the calculator above turns your call volume into a monthly bill.
How Grok Voice 3.0 is billed
| Item | List price |
|---|---|
Speech-to-speech session, grok-voice-3.0 | $0.08 per minute ($4.80 an hour) |
| Text input sent into a session | $0.004 per text input |
grok-voice-2.0 (deprecated) | same as 3.0 |
| Web search tool, from xAI's general Tools pricing | $5 per 1,000 calls |
| Collections search tool, from xAI's general Tools pricing | $2.50 per 1,000 calls |
Source: docs.x.ai/developers/pricing, read 19 Oct 2026. The tool prices come from the page's Tools Pricing table, which is written for tool use in general; the Voice Pricing table does not say whether voice sessions bill them the same way. That is why searches are a separate line you can set to zero.
A few things do not change the price: the voice you pick (all 28 built-in voices cost the same), the language, reasoning on or off (reasoning.effort), and the audio format.
Worked examples
These are the three presets in the calculator. Call lengths and text inputs are our assumptions, not xAI figures; change them to match your own logs.
| Scenario | Volume | Minutes per call | Monthly cost | Per call |
|---|---|---|---|---|
| Small business line | 300 calls a month | 3 | $73.20 | $0.244 |
| Support team (1 web search per call) | 200 calls a day | 5 | $2,502 | $0.417 |
| Starlink-sized line | 20,000 calls a day | 4 | $196,800 | $0.328 |
The last row borrows the one volume figure xAI has published: in the Grok Voice 3.0 launch post it says Starlink call centers answer 20,000 customer calls a day with Grok Voice, about 70% without a human. xAI does not publish the average call length, so the 4 minutes is ours.
What moves the bill
- Minutes, almost entirely. At $0.08 a minute, the text-input fee only matters if you push many items per call: ten tool results cost $0.04, half a minute of talk.
- Silence is still session time. You pay per minute of session, so hang up when the conversation ends instead of leaving the socket open. If you use
turn_detection.idle_timeout_msto re-engage quiet callers, give the agent a rule for when to end the call. - Shorter answers mean shorter calls. xAI's prompting guide suggests 1–2 short sentences per turn for voice agents; that also trims minutes. The Grok Voice 3.0 prompt generator writes a system prompt in that shape.
- Phone lines have their own costs. Calls that come in over SIP still pay the per-minute rate to xAI; your carrier or SIP trunk bills its own minutes separately.
How it compares, by xAI's numbers
xAI's launch post puts Grok Voice 3.0 at $0.025 per task, against $0.030 for OpenAI's GPT-Live-1 and $0.055 for GPT-Realtime-2.1 on the same benchmark. Those are the vendor's own figures on a benchmark it chose; they are useful as a rough ratio, not as your bill. For the text to speech, transcription and Voice 3.0 rates side by side, see the Grok Voice API pricing calculator.
Next steps
When the numbers work, the Grok Voice 3.0 session builder writes the session.update config and starter code for your agent, the Grok voice agent guide walks through the build, and the Grok Voice 3.0 guide covers what the model is and how it scores. To hear the voices first, use the free Grok text to speech tool.
Last updated: