Grok Voice: Voice Agents and 21 New Voices
On this page
Updated November 2, 2026 — voice. Grok Voice 3.0 (October 14, 2026) replaces deprecated Grok Voice 2.0 at the same price of $0.08 per minute; the alias
grok-voice-latestpoints to 3.0. For transcription,grok-voice-transcribe-2.0is the default model; 1.0 reached end of life on October 2, 2026. Supplement November 3, 2026: The table below is updated (voice agent at $4.80 per hour instead of $3 recorded on July 28), and the catalog now includes 28 built-in voices. Sources: Pricing · Text to Speech.
xAI’s voice division was long the quiet sibling of the lineup: a solid, documented API, but effectively reserved for teams capable of development. Early July 2026, two successive announcements changed this. They deserve to be read together because they outline two distinct entry points into the same product.
Two announcements, two audiences
The first arrived on July 1: the Voice Agent Builder. The tool promises creation of a custom voice agent “in under 2 minutes, without a line of code”. In practice, you define the agent’s role, choose its voice, provide its knowledge, then deploy it — all from an interface, without writing code. This is precisely the audience missing from xAI’s voice offering: the business manager who wants to prototype a hotline or automated switchboard without tying up a development team. The API does not disappear; it remains the path for deep integrations, where the voice agent must converse with your CRM or ticketing system.
Five days later, on July 6, xAI announced 21 new multilingual “flagship” voices for Grok Voice, and improved the naturalness of the original five along the way. For you, this multilingual expansion matters: a voice agent in English is only credible if the voice itself is. Before locking in a choice for a project, take time to listen to the voices available in your target language — output varies significantly from one voice to another, and changing voices after launching a customer service is always harder than choosing well upfront.
What the Voice API costs
The pricing grid published on docs.x.ai (recorded November 3, 2026) fits in four lines, and it is straightforward tabular data:
| Component | Price |
|---|---|
| Voice agent (real-time, grok-voice-3.0) | $0.08 per minute, or $4.80 per hour, plus $0.004 per text input |
| Text-to-speech (TTS) | $15 per million characters |
| Batch transcription (STT batch) | $0.10 per hour |
| Streaming transcription (STT streaming) | $0.20 per hour |
The most interesting line is the first, and not just for the amount. The voice agent is billed per hour, not per token. The direct consequence: the budget for a call center or hotline is calculated quickly. If your service handles 200 hours of conversation per month, you know the bill will be $960 — regardless of exchange verbosity, regardless of response complexity. This is a cost model finance teams understand immediately, whereas token-based billing requires always-fragile projections.
The gap between the two transcription modes tells a story too. Batch transcription costs $0.10 per hour, streaming transcription $0.20: real-time costs exactly double. This is the classic industry pattern, and it calls for a simple reflex: only pay for streaming where immediacy truly serves the use case. Live meeting captions justify real-time; transcribing call recordings for analysis after the fact never does — batch does the same job for half the price.
What this changes for your existing knowledge
If you took our “Grok Voice & Audio”, “Voice Agent API”, and “TTS API” courses, nothing you learned there is obsolete. The implementation mechanics — real-time WebSocket, voice activity detection (VAD), audio formats — remain exactly as described in those courses. What’s added is a layer on top: a no-code tool for fast prototyping, a significantly expanded voice catalog, and the pricing grid above, updated, which lets you cost a project before writing the first line.
In practice, the right approach for a team new to the subject now consists of prototyping in the Voice Agent Builder — two minutes is enough to hear what your use case would sound like — then, if the prototype convinces, switching to the API for real integration. The reverse order, starting with code, no longer makes sense to validate an idea.
Official source: docs.x.ai/developers/models and x.ai/news
Key points to remember
- Voice Agent Builder: create a voice agent without code (July 1)
- 21 new multilingual voices announced July 6; 28 voices in catalog as of November 3
- Voice agent: $4.80 per hour (recorded November 3; $3 at July 28 recording) — a budget calculated in time, not tokens
- Transcription: $0.10/hr batch, $0.20/hr streaming — pay for real-time wisely