The cheapest transcription API in my price set is Groq's hosted Whisper Large v3 Turbo at $0.04 per hour. Hosted open-weights Whisper is the cheapest group overall. Live and real-time tiers cost the most: Azure real-time is $1.00 per hour, and OpenAI's live transcription is $0.017 per minute. All prices are list prices as of 2026-10-11, in USD.
How I compare prices
Vendors list prices either per minute or per hour. This article shows each price in the unit the vendor publishes, with no conversion, so every number matches the official price page. I do not add taxes, minimum commitments or volume discounts, because those vary by contract.
I did not run STT benchmarks for this article. The numbers come from official price pages, so they show what each vendor charges, not how accurate each model is. For how I handle sourcing on LLM work, see my methodology. To estimate costs for a mixed audio and text workload, use the calculator.
Prices listed per hour
| Model | Vendor | Price per hour | Basis |
|---|---|---|---|
| Whisper Large v3 Turbo | Groq | $0.04 | Listed per hour |
| Whisper Large v3 | Groq | $0.111 | Listed per hour |
| Melia 1 | Speechmatics | $0.12 | Pre-recorded |
| Universal-2 | AssemblyAI | $0.15 | Pre-recorded, pay as you go |
| Azure Speech to Text (batch) | Microsoft Azure | $0.18 | S1 batch, East US |
| Reverb Transcription | Rev AI | $0.20 | English |
| Universal-3.5 Pro | AssemblyAI | $0.21 | Pre-recorded, pay as you go |
| Scribe v2 | ElevenLabs | $0.22 | Batch |
| Speechmatics Enhanced | Speechmatics | $0.38 | Pre-recorded |
| MiniMax Speech-to-Text | MiniMax | $0.38 | Listed per hour |
| Ink-2 | Cartesia | $0.39 | Scale plan effective rate |
| Azure Speech to Text (real-time) | Microsoft Azure | $1.00 | S1 standard real-time, East US |
Prices listed per minute
| Model | Vendor | Price per minute | Basis |
|---|---|---|---|
| Whisper Large v3 | Together AI | $0.0015 | Batch |
| gpt-4o-mini-transcribe | OpenAI | $0.003 | File transcription (estimated) |
| Speech-to-Text V2 dynamic batch | Google Cloud | $0.003 | Dynamic batch recognition |
| Nova-3 (monolingual) | Deepgram | $0.0043 | Pre-recorded, pay as you go |
| gpt-transcribe | OpenAI | $0.0045 | File transcription |
| Whisper Large | Rev AI | $0.005 | English |
| Nova-3 (multilingual) | Deepgram | $0.0052 | Pre-recorded, pay as you go |
| gpt-4o-transcribe | OpenAI | $0.006 | File transcription (estimated) |
| Whisper (whisper-1) | OpenAI | $0.006 | File transcription |
| Scribe v2 | ElevenLabs via fal.ai | $0.008 | Per input audio minute |
| Speech-to-Text V2 (Chirp / standard) | Google Cloud | $0.016 | 0-500,000 min/month |
| gpt-live-transcribe | OpenAI | $0.017 | Live transcription |
Rows marked "estimated" are flagged that way on the price source.
Three points stand out:
- The cheapest options are hosted open-weights Whisper. Groq's Turbo is the lowest per-hour price, and Together AI's Whisper Large v3 is the lowest per-minute price at $0.0015.
- OpenAI's own file models sit in the middle. gpt-4o-mini-transcribe is $0.003 per minute, gpt-transcribe is $0.0045, and whisper-1 is $0.006.
- Live and real-time pricing costs the most. OpenAI's live transcription is $0.017 per minute, and Azure's real-time tier is $1.00 per hour.
Batch vs streaming
Where a vendor lists both modes, the gap is large. Each row compares prices in the same unit.
| Vendor | Batch / file | Live / standard |
|---|---|---|
| Azure | $0.18 per hour | $1.00 per hour (real-time) |
| OpenAI | $0.0045 per minute (gpt-transcribe) | $0.017 per minute (gpt-live-transcribe) |
| Google Cloud | $0.003 per minute (dynamic batch) | $0.016 per minute (standard V2) |
Google's standard tier is not strictly a streaming price, but it is the full-rate alternative to dynamic batch. If your workload can wait, batch or pre-recorded pricing saves the most.
What the listed price leaves out
The listed price is only a starting point. Before you choose a vendor, check:
- Language coverage. Rev's Reverb and Whisper Large prices are for English. Deepgram's Nova-3 multilingual costs more than monolingual ($0.0052 vs $0.0043 per minute).
- Tier limits. Google's $0.016 per minute applies to the 0-500,000 minutes per month tier. Prices beyond that tier are not in my data.
- Billing unit. The fal.ai Scribe v2 price is per input audio minute. I have no data on rounding.
- Who serves the model. Whisper on Groq, Together AI or Rev AI is open-weights Whisper run by a third party, so the price covers inference only. Accuracy on your audio may differ between hosts, so test it.
FAQ
Which transcription API is cheapest?
In my set, Groq's Whisper Large v3 Turbo is the cheapest per-hour price at $0.04 per hour. Together AI's Whisper Large v3 is the cheapest per-minute price at $0.0015. Both are hosted open-weights models, so test accuracy on your own audio.
How much more does live transcription cost than batch?
Azure real-time is $1.00 per hour versus $0.18 for batch. OpenAI live transcription is $0.017 per minute versus $0.0045 for gpt-transcribe file transcription.
Does the listed price include everything I will pay?
No. These are list prices for audio processing only. They exclude taxes, discounts and any add-ons billed separately, so check each vendor's full price page before budgeting.