xAI's speech-to-text API costs $0.10 an hour in batch, and moved to 2.0 on 2 October
xAI's Speech to Text is a transcription API covering 25 languages[1]. It works in two modes: batch, where a recording is sent as a whole, and streaming, where audio is sent as it plays[1].
It costs $0.10 per hour of audio in batch and $0.20 per hour in streaming[2]. In the same Voice API, Speech to Speech for spoken conversation costs $0.08 per minute, and Text to Speech costs $15 per million characters[2].
grok-voice-transcribe-2.0 was added on 17 September[1]. It is also the model used when none is specified[1].
grok-voice-transcribe-1.0 reached end of life on 2 October[1]. Requests that name it are served by 2.0 at the same price[1]. According to the release notes, 2.0 is more accurate[1].
vad_threshold is the threshold of the voice activity detection that skips audio without speech[1]. It can be set in both batch and streaming[1]. Lower values make quieter or noisier speech, such as narrowband telephone audio, easier to transcribe, and 0 turns the detection off[1].
Streaming also supports Smart Turn[1]. With smart_turn enabled, a machine learning model predicts at each silence whether the speaker has finished, which reduces false cuts during dictation, number sequences and pauses mid-sentence[1]. smart_turn_timeout sets the longest silence allowed before the turn ends anyway[1].