xAI Releases Grok STT 1.0: Speaker Diarization, 25+ Languages, $0.10 per Hour
xAI released Grok STT 1.0 on July 23, a dedicated speech to text api available via the xAI REST API at the /v1/stt endpoint.
The model transcribes audio files to text across more than 25 languages and includes features aimed at making transcripts usable without post-processing: word-level timestamps for precise alignment, speaker diarization to label who said what in multi-participant audio, multichannel audio support, and intelligent inverse text normalization. An adjustable Voice Activity Detection threshold was also added the same day, giving developers control over how the model handles noise and silence at the edges of speech.
Pricing is flat at $0.10 per hour of audio — straightforward compared to per-token billing on text models, and competitive with the established players in the transcription market like AssemblyAI and Deepgram.
This is a distinct release from xAI's earlier audio work. The company launched separate Text-to-Speech and Speech-to-Text APIs in April 2026, and Grok Voice Think Fast 1.0, a voice agent model, arrived around the same time. grok stt 1.0 is the cleaned-up, standalone version of the transcription capability — a dedicated endpoint rather than a feature bundled inside a broader product.
xAI hasn't published benchmark comparisons against Whisper, Google's Speech-to-Text, or other dedicated speech recognition api configurations, so independent performance comparisons aren't yet available. At $0.10/hour it sits below several competitors on price, but whether the quality holds up on accented speech, domain-specific vocabulary, or difficult audio conditions is something developers will need to test themselves.
