What OpenAI transcription costs per hour of audio

OpenAI’s Whisper API (whisper-1) costs $0.006 a minute, which is $0.36 an hour of audio, per OpenAI’s pricing page checked 2026-09-27. The newer GPT Transcribe is cheaper at $0.0045 a minute ($0.27 an hour), and the streaming GPT Live Transcribe costs $0.017 a minute ($1.02 an hour). Files are capped at 25 MB each.

Price per minute and per hour

OpenAI’s list prices, checked 2026-09-27 on openai.com/api/pricing and the speech-to-text guide.

Model	$/min	$/hour	4 h/month
GPT Transcribe (batch)	$0.0045	$0.27	$1.08
GPT Live Transcribe (streaming)	$0.017	$1.02	$4.08
gpt-4o-transcribe	$0.006	$0.36	$1.44
gpt-4o-mini-transcribe	$0.003	$0.18	$0.72
whisper-1	$0.006	$0.36	$1.44

OpenAI now points new work at GPT Transcribe; the older models are still listed. If you only need the cheapest OpenAI option, gpt-4o-mini-transcribe is it.

You pay for the length of the audio you send. Silence counts, which is why trimming long pauses before upload saves money.

What a month of dictation costs

Four hours of speech a month is about 12 minutes of talking every workday. On GPT Transcribe that is about $1.08 a month, or $13 a year. On the streaming model it is $4.08 a month.

The streaming premium is worth a look. Nota’s recommended OpenAI pick is plain GPT Transcribe, at $0.27 an hour. GPT Transcribe (streaming) starts transcribing while you talk, but costs about four times as much per minute ($1.02 an hour), so choose it only if that head start matters more to you than the price.

For comparison, a dictation subscription is $12 to $15 a month, and a local model on your Mac costs nothing.

If a small language model then cleans up punctuation and spoken corrections, add under a cent an hour on the model Nota uses by default. The transcription is most of the bill.

The 25 MB file limit

OpenAI’s transcription endpoint rejects files over 25 MB. At typical recording quality that is about 13 minutes of audio. Short dictations never get near it. A one-hour interview does.

Three ways around it:

1. Compress. Speech does not need music quality. Mono at 64 kbps fits about 50 minutes in 25 MB. Convert to M4A or MP3 at that bitrate before sending.
2. Split. Cut the file into parts under 25 MB, transcribe each, join the text. Cut on pauses, not mid-word.
3. Use a different engine for long files. Local Whisper or Parakeet on your Mac has no upload limit. Other cloud providers accept larger files.

In Nota, cloud providers get the whole file, so the limit applies as is: an OpenAI file over 25 MB fails. The recording stays in History, so you can switch the mode to a local model and run it again. Long pauses are trimmed first, which helps, and local models take long files without a limit. The Transcribe docs list the supported formats.

Cheaper and free alternatives

• Groq’s hosted Whisper large-v3 turbo: $0.04 an hour as of 2026-09-27, the cheapest major API we found.
• AssemblyAI Universal-2: $0.15 an hour.
• ElevenLabs Scribe v2: $0.22 an hour.
• Whisper on your Mac: $0. The same open model family, run locally with no key; large-v3 turbo is a 1.5 GB download.

The full table for 11 providers is in speech-to-text API prices per hour.

To use OpenAI in Nota, paste your key in the OpenAI panel and press Verify; setup is in the cloud models docs.