How much does it cost to transcribe an hour of audio?
At list prices read on 2026-10-25, transcribing one hour of audio costs about $0.15 to $0.36 on the main speech-to-text APIs. AssemblyAI Universal-2 is $0.15 an hour, OpenAI's gpt-4o-mini-transcribe $0.18, Deepgram Nova-3 $0.26, and OpenAI whisper-1 and Amazon Transcribe $0.36. Google Speech-to-Text V2 is $0.96 an hour at its standard rate and $0.18 in dynamic batch mode. Sending the audio to Gemini 2.5 Flash-Lite as audio tokens comes to roughly $0.04 an hour.
What does each speech-to-text API charge per hour?
First pricing tier, US region, pay as you go. We read each provider's pricing page on 2026-10-25; prices change, so check the page before you budget.
| Service | Published rate | Per hour | Per 1,000 hours |
|---|---|---|---|
| Gemini 2.5 Flash-Lite (audio tokens, estimate) | $0.30 / 1M audio input tokens, $0.40 / 1M output | $0.039 | $39 |
| Gemini 2.5 Flash (audio tokens, estimate) | $1.00 / 1M audio input tokens, $2.50 / 1M output | $0.145 | $145 |
| AssemblyAI Universal-2 | $0.15 / hour | $0.15 | $150 |
| OpenAI gpt-4o-mini-transcribe | $0.003 / minute | $0.18 | $180 |
| Google Speech-to-Text V2, dynamic batch | $0.003 / minute | $0.18 | $180 |
| AssemblyAI Universal-3.5 Pro | $0.21 / hour | $0.21 | $210 |
| Deepgram Nova-3, pre-recorded | $0.0043 / minute | $0.258 | $258 |
| OpenAI gpt-transcribe | $0.0045 / minute | $0.27 | $270 |
| OpenAI whisper-1 or gpt-4o-transcribe | $0.006 / minute | $0.36 | $360 |
| Amazon Transcribe, batch | $0.0001 / second | $0.36 | $360 |
| Google Speech-to-Text V2, standard | $0.016 / minute | $0.96 | $960 |
Volume tiers lower several of these. Google's standard rate drops to $0.01 a minute after 500,000 minutes a month, for example.
How is the Gemini figure calculated?
Gemini counts audio at 32 tokens a second, so one hour is 115,200 input tokens. The transcript is output: about 150 spoken words a minute is 9,000 words an hour, or roughly 12,000 tokens. On Gemini 2.5 Flash-Lite that is 115,200 x $0.30 / 1M plus 12,000 x $0.40 / 1M, about $0.039. On Gemini 2.5 Flash it is about $0.145. Google's Batch API halves both rates. The output figure is an estimate because the transcript length depends on how fast people talk and how much silence the recording holds.
A general model returns whatever text you ask for, so word-level timestamps, speaker labels and consistent formatting depend on your prompt. Dedicated speech-to-text APIs return timestamps by default and sell diarization (who said what) as a feature or a separate model.
What does it cost to transcribe a whole library?
Multiply the hourly rate by the hours of audio. A 10,000-hour archive of podcasts, calls or recorded meetings costs about $390 on Gemini 2.5 Flash-Lite, $1,500 on AssemblyAI Universal-2, $1,800 on gpt-4o-mini-transcribe, $3,600 on whisper-1 or Amazon Transcribe, and $9,600 on Google's standard rate before volume discounts. Video counts the same way: you pay for the length of its audio track.
What else adds to the transcription bill?
Where Mixpeek fits
Mixpeek reads your files where they already live, in S3, GCS or any bucket, and turns what's in them into data your software can search, classify and moderate. Its multimodal extractor transcribes each segment of audio or video with Whisper when you set
run_transcription, stores the text with its
timestamps, and embeds it, so a retriever returns
the moment something was said.Who it fits: teams that need to search, filter or classify what was said across many recordings. Who it does not: if you only need a text file of one recording, call a speech-to-text API directly.
Frequently Asked Questions
What is the cheapest way to transcribe audio?
At list prices on 2026-10-25, sending audio to Gemini 2.5 Flash-Lite is the cheapest of the options above, at about $0.04 an hour. Among dedicated speech-to-text APIs, AssemblyAI Universal-2 is cheapest at $0.15 an hour, followed by OpenAI gpt-4o-mini-transcribe and Google's dynamic batch mode at $0.18. Running open-source Whisper on your own GPU has no per-minute fee; you pay for the hardware and the time it runs.
How much does Whisper cost per hour?
OpenAI's hosted whisper-1 costs $0.006 a minute, which is $0.36 an hour. The open-source Whisper models are released under the MIT license and are free to run on your own hardware.
How many tokens is an hour of audio?
In Gemini, 115,200 tokens: 32 tokens a second for 3,600 seconds. One minute is 1,920 tokens.
How much does it cost to transcribe a podcast episode?
A one-hour episode costs $0.15 to $0.36 on most speech-to-text APIs and about $0.04 to $0.15 sent to Gemini, at the rates in the table above. A weekly one-hour show is 52 hours a year, under $20 a year at $0.36 an hour.
Is real-time transcription more expensive than batch?
Yes, usually. OpenAI's live transcription is $0.017 a minute against $0.003 to $0.006 for its batch models, and Google charges $0.016 a minute at its standard rate against $0.003 in dynamic batch, which returns results later in exchange for the lower price.