Pricing and budget · 2026-09-22
Grok Voice STT: calculating cost with per-hour pricing instead of tokens
Covers that xai/grok-voice-stt, unlike most text models, is billed per hour of processed audio rather than per token, and how to estimate the cost of a transcription workload.
The unit is hours, not tokens
While most models in this catalog are priced per million tokens, xai/grok-voice-stt is billed at $0.10 per hour of processed audio. That's a cost model directly tied to the input audio file's length — the content of the speech, word density, or the output text's length don't affect the bill, only the audio recording's duration does.
The model is positioned on its catalog page for transcription, note capture, and audio-input product flows, and carries a 'Beta' status label; it's worth testing accuracy with your own audio samples (accents, background noise, technical terminology) before going to production.
Calculating a transcription workload's cost
Since it's a flat per-hour price, the math is simple: estimate the total audio hours you'll process daily or monthly and multiply by $0.10. For example, a product processing an average of 50 hours of meeting recordings a day can expect around $5 a day, roughly $150 a month in STT cost — that estimate is independent of speaker count or language.
This per-hour model simplifies budgeting compared to a token-based text model: you don't need to know in advance how long the output text will be, just the input audio duration.
- Cost = total audio hours to process × $0.10.
- Output text length or speaker count doesn't affect the bill.
- Budgeting requires knowing only the input audio duration.
Factor the Beta status into your production plan
Since the model is listed with a 'Beta' label, its behavior can change over time. For a critical transcription flow (a legal or medical recording, for example), periodically retesting accuracy and keeping a fallback plan to switch to an alternative STT model if needed raises production reliability.
Frequently asked questions
Does the output text's length affect the bill?
No, billing is based solely on processed input audio duration (per hour); the output text's length doesn't affect cost.
Is there a minimum charge for very short audio clips (a few seconds)?
The model page states a per-hour unit price; check the current pricing page or API documentation for exact rounding or minimum-charge behavior on very short clips.