Cheapest speech-to-text API
At 100 hours a month, the cheapest for recorded audio is Groq's Whisper Large V3 Turbo at $0.04 an hour, and for live audio is AssemblyAI's Universal-Streaming at $0.15.
Cheapest for recorded audio
The cheapest speech-to-text API for recorded audio is Groq's Whisper Large V3 Turbo, at $0.04 an hour ($0.0007 a minute), 67% less than the next cheapest, Speechmatics' Melia 1 at $0.12. Prices are regular pay-as-you-go rates at 100 hours a month.
| # | Vendor and option | Per month | Per hour | Per minute |
|---|---|---|---|---|
| 1 | Groq Whisper Large V3 Turbo | $4 | $0.04 | $0.0007 |
| 2 | Speechmatics Melia 1 | $12 | $0.12 | $0.002 |
| 3 | AssemblyAI Universal-2 | $15 | $0.15 | $0.0025 |
| 4 | OpenAI gpt-4o-mini-transcribe | $18 | $0.18 | $0.003 |
| 4 | Google Cloud V2 dynamic batch | $18 | $0.18 | $0.003 |
| 4 | Azure AI Speech Batch transcription | $18 | $0.18 | $0.003 |
| 7 | ElevenLabs Scribe v2 | $19.10 | $0.191 | $0.0032 |
| 8 | Rev AI Reverb (English) | $20 | $0.20 | $0.0033 |
| 9 | Deepgram Nova-3 Monolingual | $25.80 | $0.258 | $0.0043 |
| 10 | Gemini API Gemini 3.5 Transcribe | $30.60 | $0.306 | $0.0051 |
| 11 | Amazon Transcribe Standard | $36 | $0.36 | $0.006 |
The order shifts at 10 hours a month, where free monthly minutes or volume tiers kick in; the calculator shows each.
Cheapest for live audio
For live audio, the cheapest is AssemblyAI's Universal-Streaming, at $0.15 an hour ($0.0025 a minute), 25% less than the next cheapest, Speechmatics' Melia 1 at $0.20.
| # | Vendor and option | Per month | Per hour | Per minute |
|---|---|---|---|---|
| 1 | AssemblyAI Universal-Streaming | $15 | $0.15 | $0.0025 |
| 2 | Speechmatics Melia 1 | $20 | $0.20 | $0.0033 |
| 3 | ElevenLabs Scribe v2 Realtime | $21.45 | $0.214 | $0.0036 |
| 4 | Deepgram Nova-3 Monolingual | $46.20 | $0.462 | $0.0077 |
| 5 | Gemini API Gemini 3.5 Transcribe Live | $53.55 | $0.536 | $0.0089 |
| 6 | Amazon Transcribe Standard | $60 | $0.60 | $0.01 |
| 7 | Google Cloud V1 with data logging | $95.04 | $0.95 | $0.0158 |
| 8 | Azure AI Speech Real-time transcription | $100 | $1 | $0.0167 |
| 9 | OpenAI gpt-live-transcribe | $102 | $1.02 | $0.017 |
The order shifts at 10 hours a month, where free monthly minutes or volume tiers kick in. Groq and Rev AI list no streaming price.
How the cheapest bill
Check how the cheapest bill before choosing on price alone:
- Groq
- Every request is billed for at least 10 seconds.
- AssemblyAI
- Pre-recorded audio is billed per hour of audio submitted; streaming is billed for the time the connection is open, idle time included. AssemblyAI doesn't state a rounding increment.
Cheapest with speaker labels
Telling speakers apart costs extra on some APIs. Each vendor's cheapest recorded-audio option with labels added:
| Vendor and option | Speaker labels | Per hour with labels |
|---|---|---|
| Speechmatics Melia 1 | Included | $0.12 |
| AssemblyAI Universal-2 | +$0.02 an hour | $0.17 |
| Azure AI Speech Batch transcription | Included | $0.18 |
| Deepgram Nova-3 Monolingual | Included | $0.258 |
| Gemini API Gemini 3.5 Transcribe (estimate) | Included | $0.306 |
| OpenAI gpt-4o-transcribe-diarize (estimate) | Separate model | $0.36 |
| Amazon Transcribe Standard | Included | $0.36 |
Google Cloud, ElevenLabs, Groq and Rev AI don't price or mention speaker labels on their pricing pages, so they're left out.
Cheapest way to start: free credits
| Vendor | Credit | About this much audio |
|---|---|---|
| Google Cloud | New Google Cloud customers get $300 of free credit to try any Google Cloud product. | 1,667 hours on V2 dynamic batch |
| Speechmatics | New accounts get $100 of free credit, no card required. | 833 hours on Melia 1 |
| Deepgram | Every new account gets $200 of free credit, which doesn't expire. | 775 hours on Nova-3 Monolingual |
| AssemblyAI | Sign-up comes with $50 of free credit and no card required; AssemblyAI says that covers up to 185 hours of pre-recorded or 333 hours of streaming transcription. | 185 hours vendor's figure |
| Rev AI | New accounts get free credits worth 5 hours of Reverb transcription, usable on any product. | 5 hours vendor's figure |
Hours we work out use the vendor's cheapest recorded-audio price; Google's credit is for any Google Cloud product. Every free tier is compared on free speech-to-text APIs.
Free allowances that recur
- Google Cloud: The V1 API's first 60 minutes a month are free; the V2 table lists no free minutes.
- Gemini API: The free tier charges nothing for input or output, but Google may use free-tier content to improve its products, which it doesn't on the paid tier.
- Amazon Transcribe: New accounts get 60 minutes a month free, for the first 12 months only.
- Azure AI Speech: A separate free tier (F0) gives 5 audio hours a month of real-time transcription; batch isn't supported on it.
- ElevenLabs: The Free / pay-as-you-go column of the API price table includes 4 hours 30 minutes of Scribe v2 and 2 hours 30 minutes of Scribe v2 Realtime.
- Groq: The free plan transcribes up to 7,200 audio seconds an hour and 28,800 a day (8 hours) per Whisper model at no charge, within its rate limits.
Promotions that undercut these prices
- Azure AI Speech MAI Transcribe-2 (batch): $0.0017 a minute ($0.10 an hour), the only price published, until 31 December 2026.
- Deepgram Nova-3 Monolingual (streaming): $0.0048 a minute ($0.288 an hour) instead of $0.462, with no end date given.
- Deepgram Flux English (streaming): $0.0065 a minute ($0.39 an hour) instead of $0.462, with no end date given.
- Deepgram Nova-3 Multilingual (streaming): $0.0058 a minute ($0.348 an hour) instead of $0.552, with no end date given.
- Speechmatics Linden 1 (voice agents) (streaming): $0.0025 a minute ($0.15 an hour) instead of $0.20, with no end date given.
Price your own
- Speech-to-text cost calculatorYour own volume, with billing increments for short clips and every model
- What it costs to transcribe an hour of audioOne recording
Prices on this page were checked against the official pricing pages on .