xAI Launches Grok STT and TTS APIs

xAI has unveiled two standalone audio APIs, Grok Speech to Text (STT) and Grok Text to Speech (TTS). Built on the same stack used for Grok Voice, Tesla vehicles, and Starlink customer support, they're designed to plug directly into voice agents, real-time transcription, accessibility, podcasts, and interactive audio experiences.

STT transcribes large audio files in milliseconds via a REST API, while real-time processing is served through a WebSocket API built for the lowest possible latency. It includes word-level timestamps, speaker diarization, and multichannel support, along with Inverse Text Normalization that contextually normalizes numbers, dates, currencies, and more.

On the performance side, xAI presented comparisons against commercial models across phone calls, meetings, video/podcasts, and telephony. The table shows an overall Word Error Rate (WER) of 6.9% for Grok STT, and 5.0% on phone-call entity recognition, which xAI claims is lower than ElevenLabs, Deepgram, and AssemblyAI. It also stated support for 25+ languages and the ability to handle speaker diarization on both real-time and recorded audio.

TTS aims for natural, expressive speech, offering a REST API for long-text conversion and a WebSocket API for real-time generation. Speech tags like [laugh], [sigh], [whisper], , , and allow fine-grained control over intonation and emotion, priced at $4.20 per 1 million characters.

Pricing for STT is $0.10/hour for batch and $0.20/hour for streaming, with rate limits and further details available in the xAI API console. Through this launch, xAI is positioning the offering as a package that products needing both voice input and output can integrate directly, without building separate voice infrastructure.