Most accurate
A5S Async
$0.18/ audio hour
- Transcripts of recorded audio
- Speaker labels included
- Word timestamps and subtitles
- Uploads, URLs and webhooks
- Names from enrolled voices
Pay per audio hour.
No subscription and no minimum term. Add credits as you go, and pay only for the audio you send, to the millisecond.
5 free hours on every account, across all models
A5S Streaming
$0.30/ audio hour
- Live transcripts over WebSocket
- Partial and final results
- Optional live speaker labels
- For live captions and voice agents
Diarization
$0.08/ audio hour
- Speaker turns without a transcript
- Finds the number of speakers itself
- Works from the voice, not the words
- JSON segments or RTTM
Enabled per account on request
Speaker ID
$0.02/ audio hour
- Enroll voices, then recognize them
- Accurate through noise and music
- Billed on enrollment and identification audio
- Included in transcripts and live streams
Enabled per account on request
What will it cost?
Model
More accuracy per dollar.
No model makes fewer errors on real meetings. The two that cost less make 1.9 and 2.7 times as many.
| Provider | Model | Price per audio hour | Meeting error rate (cpWER, lower is better) |
|---|
List prices with speaker labels, November 2, 2026. Microsoft's $0.10 is introductory until December 31, 2026; AssemblyAI is $0.21 plus $0.02 for diarization; Google prices by token ($0.31 is its own estimate). Error rate: macro cpWER on NOTSOFAR-1, AMI, CHiME-6, AISHELL-4 and AliMeeting. Method
Thousands of hours a month?
We'll price for your volume, run an evaluation on your own audio, and set up the deployment you need.
- Volume pricing from 10,000 hours a month
- Higher limits and concurrency
- Private cloud (VPC) or on-premise deployment
- An evaluation on a sample of your audio
- Access to Diarization and Speaker ID
Questions.
How is usage billed?
By the length of the audio each model processes, to the millisecond, at the hourly rate above. Your first 5 hours are free on every account and shared across all models.
How do I pay?
Add prepaid credits in the console, from $5. Turn on auto refill to top up automatically when your balance runs low.
Are speaker labels extra?
No. Speaker labels are included in A5S Async, and live speaker labels are included in A5S Streaming. Names from enrolled voices are included in transcripts and live streams; the standalone Speaker ID API is billed on its own.
Which languages are supported?
A5S Async and A5S Streaming transcribe English. Diarization and Speaker ID work from the voice, not the words.
Do you train on my data?
No. See Data and privacy in the docs.
Can I run AirCaps in my own cloud?
Yes. Private cloud (VPC) and on-premise deployments are available for volume customers. Talk to sales.
Hear what other models miss.
Run AirCaps on your own audio. Every account includes 5 free hours.