xAI Grok Voice Adds 21 New Flagship Voices, Supporting Korean and 25+ Languages

What was added

On July 6, xAI announced 21 new flagship voices for Grok Voice through its official newsroom. The structure adds a new lineup, including Lumen, Castor, Naksh, Atlas, Carina, Zagan, Helix, Orion, and Luna, alongside the five existing voices.

These are not just simple sample voices. xAI explained that the new voices can be used immediately in the Realtime Voice Agent API, Text to Speech API, and Grok Voice Agent Builder. In other words, developers can choose the same voices not only when assembling a separate voice chatbot, but also inside a voice agent builder that is closer to no-code.

Why this is worth viewing as a follow-up to Voice Agent Builder

On July 4, Game Sunakku introduced xAI Voice Agent Builder as a “workbench for making phone AI support agents.” This news is less about the workbench itself and more about an update that adds more voice materials users can actually choose inside that workbench.

For voice AI to enter real services such as phone support, reservations, customer support, education, and character-style guidance, “what it can say” matters, but so does “what voice it uses to say it.” That is also why xAI introduced the new voices by role. By showing use cases such as Support, Characters, Commentary, Advertising, and Education, the direction appears to be making voice AI into a product option rather than just a demo.

What multilingual support, including Korean, means

xAI stated that all of the new voices support more than 25 Grok Voice languages, and the example language list includes Korean. This is an important change for global services or customer support that mixes multiple languages.

Of course, “supported” is not the same as “natural in every situation.” Real testing is needed for accents, proper nouns, fast speech, emotional expression, and environments with poor call quality. Still, based on the official announcement, it has become clearer that Grok Voice is expanding from an English-centered voice demo into multilingual voice agent material.

The five existing voices have also improved

xAI explained that it did not only add new voices; it also refined the five existing voices, Ara, Eve, Leo, Rex, and Sal, with a new training approach. Based on the announcement’s wording, the direction is toward more natural pacing, sentence breaks, and emphasis.

xAI also says speaking tags such as [pause] and <whisper> can be used to fine-tune voice delivery. Put simply, this is not just “TTS that reads text aloud,” but something that can add direction such as pausing briefly like a support agent, speaking quietly, or emphasizing specific parts.

Points to watch before using it in a service

The more natural voice AI becomes, the larger its scope of responsibility gets. When attaching it to areas that can change real human behavior, such as customer support or reservations, services first need to set rules for AI disclosure, recording and transcript retention, handoff to human agents, and standards for handling sensitive information.

In particular, xAI is also presenting a flow in which a custom voice can be created with about one minute of audio. This feature can be attractive for brand voices or character-style guidance, but services need to be careful about voice cloning without consent and the possibility of confusion. Costs may also vary depending on per-minute usage, so before real operation, the latest xAI pricing table and expected call volume need to be calculated together.

In short, this announcement does not end at “21 more voices were added.” It is more accurate to view it as a sign that voice AI is moving beyond a fun in-app feature and preparing the materials needed for multilingual customer support, education, and character-style services.