Model Details
Released
May 7, 2026
Context Length
16,000
Max Output
2,000
Where is gpt-realtime-whisper a perfect fit?
GPT-Realtime-Whisper is OpenAI's dedicated streaming speech-to-text model, designed to return low-latency transcript updates while someone is still speaking. Developers can tune the trade-off between transcription latency and accuracy.
- Live captions and subtitles for calls, meetings, and broadcasts
- Realtime transcription while a user is actively speaking
- Voice assistants that need text transcripts before downstream processing
- Contact-center and customer-support call transcription
- Meeting notes and live conversation logging
- Accessibility applications providing immediate speech-to-text
- Applications requiring adjustable latency versus transcription accuracy
- Streaming audio workflows where waiting for a complete recording is impractical
- Live captions and subtitles for calls, meetings, and broadcasts
- Realtime transcription while a user is actively speaking
- Voice assistants that need text transcripts before downstream processing
- Contact-center and customer-support call transcription
- Meeting notes and live conversation logging
- Accessibility applications providing immediate speech-to-text
- Applications requiring adjustable latency versus transcription accuracy
- Streaming audio workflows where waiting for a complete recording is impractical
Quick Model Estimate
Your GPT-5 Cost Estimate
💰 Total Cost
USD 3.00
for 1000 input + 1000 output tokens
📥 Input (1000 × $N/A)
USD 1.5000
📤 Output (1000 × $0.017000)
USD 1.5000
Cost Breakdown
📥 Input 50%
📤 Output 50%
Prices updated daily from official provider data.
Pricing
|
Provider
↕
|
Modality
↕
|
Service tier
↕
|
Input Price
(per 1M tokens) ↕ |
Output Price
(per 1M tokens) ↕ |
Context Size
↕
|
View
|
|---|---|---|---|---|---|---|
 OpenAI | Audio | Standard | N/A | $0.0170 | 16,000 tokens | → |
Other Models in the gpt-realtime Family
FAQs about gpt-realtime-whisper
What is gpt-realtime-whisper's context window?
gpt-realtime-whisper has a context window of 16,000 tokens, and returns up to 2,000 tokens in a single response.