Getting started
Models
The Whisper VoiceKit model catalog: pre-recorded, realtime, and voice agent models with rates and guidance.
The API exposes one branded model family, Whisper VoiceKit, across three product areas: pre-recorded transcription, realtime streaming, and voice agents. Every request selects a model with the model field, and each model is metered at its own rate.
The catalog is also available programmatically, including live rates, so your integration never needs hardcoded pricing:
/v2/modelsList all models with unit types and rates/v2/pricingSame catalog, aliased for billing pagesPre-recorded models
Pre-recorded models transcribe uploaded files and audio URLs asynchronously through POST /v2/transcripts. Both return full text, word timings, confidence scores, optional speaker labels, and SRT or VTT subtitles, plus optional speech-intelligence results (summary, chapters, entities, sentiment_results, content_safety) when the matching request flags are set. Completion can be consumed by polling or by registering a webhook_url on the create request.
Whisper VoiceKit Best
The flagship transcription model. It delivers the highest accuracy in the catalog and is tuned for a focused set of six core languages, making it the default choice for production transcripts, subtitles, and downstream AI analysis. $0.252 per audio hour, model id whisper-voicekit-best.
Whisper VoiceKit Fast
A cost-effective model with broad coverage across the full language catalog. Choose it for high-volume pipelines, long-tail languages, and workloads where price per hour matters more than the last point of accuracy. $0.18 per audio hour, model id whisper-voicekit-fast.
Realtime models
Realtime models transcribe live audio over a WebSocket. Your server mints a short-lived session token with POST /v2/realtime/token, the client connects to wss://api.realtimevoicekit.com/v2/realtime/ws with that token, streams PCM audio, and receives partial and final transcripts as JSON events. Sessions are metered per second of connection time and each connection runs for up to 55 minutes before the API asks the client to reconnect.
Whisper VoiceKit Realtime Pro
The low-latency streaming model for production voice products: live captions, dictation, meeting transcription, and agent front-ends. $0.54 per session hour, model id whisper-voicekit-realtime-pro. This is the default model for realtime sessions.
Whisper VoiceKit Realtime
A cost-effective streaming model for high-volume or budget-sensitive realtime workloads where a little extra latency headroom is acceptable. $0.18 per session hour, model id whisper-voicekit-realtime.
Voice agent model
Whisper VoiceKit Agent
Runs full speech-to-speech agent sessions: the model listens, reasons, calls your tools, and answers with audio. Mint a session token with POST /v2/voice-agent/token, connect to wss://api.realtimevoicekit.com/v2/voice-agent/ws, configure the session with a session.update message, then stream audio both ways. $5.40 per session hour, model id whisper-voicekit-agent. See the Voice Agent API docs for the full event protocol.
Medical Mode
Medical Mode is an add-on for clinical audio: set medical_mode: true on a transcript or realtime token request and dictations, consultations, and case notes are transcribed with a vocabulary tuned for healthcare. It is available in English, Spanish, German, and French at no extra charge; for other languages the flag is ignored and standard transcription runs instead.
Choosing the right model
- Accuracy first, recorded audio: choose Whisper VoiceKit Best. Interviews, podcasts, legal and medical records, anything you will publish or analyze.
- Volume first, recorded audio: choose Whisper VoiceKit Fast. Bulk archives, long-tail languages, drafts, and cost-sensitive pipelines.
- Live audio in production: choose Whisper VoiceKit Realtime Pro for the lowest latency and best live accuracy.
- Live audio at scale: choose Whisper VoiceKit Realtime when per-hour cost matters more than minimum latency.
- Conversational agents: choose Whisper VoiceKit Agent when the API should also reason and speak, not just transcribe.
whisper-voicekit-best for files and whisper-voicekit-realtime-pro for live audio. They are the defaults when model is omitted on their respective endpoints.Pricing
| Model | Unit | Rate |
|---|---|---|
| Whisper VoiceKit Best | Audio hour | $0.252 |
| Whisper VoiceKit Fast | Audio hour | $0.18 |
| Whisper VoiceKit Realtime Pro | Session hour | $0.54 |
| Whisper VoiceKit Realtime | Session hour | $0.18 |
| Whisper VoiceKit Agent | Session hour | $5.40 |
Usage is measured per second and priced at the hourly rate, with each usage event rounded up to the next cent. Your $10 free credit is applied automatically before anything is billed; see Billing and pricing for the full mechanics.
Next steps
- Transcribe your first file with Whisper VoiceKit Best.
- Stream live audio with a realtime session token.
- Build a voice agent on the agent WebSocket.
- Review billing to estimate cost for your workload.