Speech-to-text

OpenAI

OpenAI's transcription API (gpt-transcribe, gpt-4o-transcribe, whisper-1), billed per audio minute or token. US company; EU residency is sales-gated.

The main argument is convenience for projects already on the OpenAI API, with gpt-transcribe at an estimated $0.27/hour and gpt-4o-mini-transcribe at $0.18/hour. Several models are billed per token, so the per-minute figures are OpenAI's estimates rather than fixed rates, and live transcription costs $1.02/hour. There is no free tier, and EU data residency needs sales approval and adds a 10% uplift on newer models, whereas AssemblyAI and Deepgram offer EU endpoints.

Pricing

At least one rate is approximate or not on a published rate card. Confirm with the vendor. Token-billed transcription uses the vendor's estimated per-minute cost, not a fixed hourly rate. How to read these estimates.

gpt-transcribe (current flagship file-transcription model) is listed at an estimated $0.0045/min ($0.27/hour). Older and cheaper options: gpt-4o-transcribe $0.006/min ($0.36/hour; $2.50 in / $10 out per 1M tokens), gpt-4o-mini-transcribe $0.003/min ($0.18/hour; $1.25/$5 per 1M tokens), whisper-1 $0.006/min ($0.36/hour). Token-billed models are priced per token, so the per-minute figure is OpenAI's estimate. Live transcription (gpt-live-transcribe, gpt-realtime-whisper) $0.017/min ($1.02/hour). No free tier. Data residency in Europe (EEA + Switzerland, audio endpoints included) requires sales approval, abuse-monitoring controls and a Modified Retention amendment, and adds a 10% uplift on models released after March 5, 2026. DPA available. The original Whisper model weights are open source (MIT), but the API models are not.

Want a European-owned provider? See European alternatives to OpenAI among european speech-to-text providers.

Compare OpenAI

• OpenAI vs Deepgram

Found a wrong rate or plan restriction? Report a correction with a current source.