xAI expands Grok Voice with 21 new flagship voices and a no-code Voice Agent Builder

xAI has released 21 new flagship voices for Grok Voice, more than quadrupling its lineup, and paired the launch with a new no-code Voice Agent Builder for assembling custom voice agents.

What's new

"Today, we're releasing 21 new flagship voices for Grok Voice, joining the original five," xAI said in the announcement. "All of them are multilingual and available now in the realtime Voice Agent API, the Text to Speech API, and the new Grok Voice Agent Builder."

Every new voice is natively multilingual, supporting Grok Voice's full set of 25-plus languages, including English, Hindi, Gujarati, Bengali, Kannada, Spanish, French, German, Italian, Portuguese, Russian, Indonesian, Japanese, Korean, and Thai. Rather than shipping a single generic voice, xAI cast each one for a specific job: customer support, character work, commentary, advertising, and education.

Developers can also shape delivery at a granular level. xAI's documentation shows speech tags like [pause] and <whisper> that let a voice pause mid-sentence, whisper a line, or add emphasis to a phrase, giving more control over pacing than a flat text-to-speech read would allow.

The original five voices — Ara, Eve, Leo, Rex, and Sal — weren't left behind. xAI says they "got an upgrade alongside the new voices," having "been retrained with an improved recipe, so pacing, phrasing, and emphasis are more natural."

To put a new voice to work, developers use the Grok Voice Agent Builder inside the xAI console to assemble a custom voice agent, or clone a voice entirely from about a minute of a person's own audio via xAI's custom-voice tooling.

Context

Voice has become a competitive front for the frontier labs beyond raw model capability. OpenAI shipped GPT-Realtime-2.1 and a faster mini variant for voice apps earlier this month, and Google, Anthropic, and others have all pushed multimodal audio features in their APIs. xAI's original five-voice Grok Voice lineup, by comparison, was a comparatively narrow offering; this release brings it closer to parity with rivals on breadth while adding developer-facing controls — the speech tags and Agent Builder — that go beyond what a basic TTS endpoint typically offers.

Why it matters

For developers building voice agents — customer support bots, in-app assistants, narration tools — voice selection and expressive control are now a real design surface, not an afterthought bolted onto a text model. By casting voices for specific use cases and giving them fine-grained delivery controls, xAI is positioning Grok Voice as infrastructure for production voice products rather than a demo feature. The retraining of the original five voices alongside the new batch also signals that xAI intends to keep investing in voice quality on an ongoing cadence, not treat it as a one-time launch.