SpaceXAI: Grok Voice TTS 1.0

SpaceXAI: Grok Voice TTS 1.0 is a text-to-speech model from xAI, listed on api14 under the speech category. It takes written text and turns it into spoken audio, which makes it a fit for any workload where you already have text and need a voice track rather than a text response. The model id is x-ai/grok-voice-tts-1.0, and it was released by the provider on 2026-05-15.

The model accepts text as its only input modality and produces speech as its only output modality. It supports automatic language detection and can speak across more than 20 languages, so you do not have to declare a language up front for mixed or unknown input. Five built-in voices are available: eve, ara, rex, sal and leo. The context window is 15,000 tokens, with a maximum output of 13,500 tokens, which bounds how much text you can hand over in a single request.

Those capabilities point to a fairly specific set of uses. Narration and voiceover for articles, documentation or product copy are the obvious ones, since you can feed the source text straight in and get audio back. Because language detection is automatic, the same integration can serve multilingual content without routing requests by locale. The five voices let you assign a consistent speaker per project, channel or character. Accessibility features such as read-aloud for long-form text, and audio versions of chat or assistant replies, also follow directly from a text-in, speech-out model with a large input allowance.

You call it through api14's OpenAI-compatible API using the model id x-ai/grok-voice-tts-1.0, so existing OpenAI-style client code works with a change of model name and base URL. The model exposes no tunable request parameters beyond what the endpoint itself defines, so there is no temperature, top_p or similar sampling surface to configure. Voice selection is the main lever you have, drawn from the five listed voices. Plan your requests around the 15,000 token context and the 13,500 token output ceiling.

If your task is text generation, classification, reasoning or anything that returns words rather than audio, this is the wrong model, since speech is its only output. It also will not help when you need to transcribe existing audio, because text is its only input. For those cases, look at a text or speech-to-text model instead. Within its scope, though, it is a straightforward choice when you need spoken output in one of the supported voices and languages.