
In this article▾
OpenAI Whisper API pricing is $0.006 a minute, or $0.36 an hour of audio, and its replacement, gpt-transcribe, costs $0.0045 a minute. OpenAI marked whisper-1 as deprecated in August 2026 and shuts it down on 26 February 2027. We read every rate on OpenAI's pricing page on 17 November 2026, then ran a real 14 minute speech through Wireflow, which runs OpenAI's Whisper model as one node next to 100+ AI models.
On Wireflow that node is listed at 1 credit a run, whatever the length of the file. A credit costs about 1.35 cents on the monthly plans, which buys a little over 2 minutes of whisper-1 or 3 minutes of gpt-transcribe, so a longer recording costs less as one run than it does by the minute on OpenAI. This guide covers every OpenAI transcription rate, what our test file cost on each, and the limits the per-minute price doesn't show.
For the opposite job of turning text into speech, our breakdown of ElevenLabs API pricing covers the rate per 1,000 characters and what a real script cost to voice.
OpenAI transcription prices on 17 November 2026
OpenAI bills its transcription models by the minute of audio you send, the usual shape of usage-based AI API pricing. Some of them are billed in tokens behind the scenes, and for those the per-minute figure on OpenAI's pricing page is its own estimate, which is why they're marked "about" below.
| Model | Price per minute | Per hour of audio | Status on 17 Nov 2026 |
|---|---|---|---|
| gpt-transcribe | $0.0045 | $0.27 | Current, recommended for recorded files |
| whisper-1 | $0.006 | $0.36 | Deprecated, shuts down 26 Feb 2027 |
| gpt-4o-transcribe | about $0.006 | about $0.36 | Deprecated, shuts down 26 Feb 2027 |
| gpt-4o-mini-transcribe | about $0.003 | about $0.18 | Deprecated, shuts down 26 Feb 2027 |
| gpt-4o-transcribe-diarize | about $0.006 | about $0.36 | Deprecated, shuts down 26 Feb 2027 |
| gpt-live-transcribe | $0.017 | $1.02 | Current, for live audio |
OpenAI's deprecation list names gpt-transcribe or gpt-live-transcribe as the way forward for all four older models, with the same 26 February 2027 shutdown date. Moving a recorded-file pipeline from whisper-1 to gpt-transcribe cuts the price of an hour of audio from $0.36 to $0.27. The lowest rate on the page belongs to gpt-4o-mini-transcribe at about $0.003 a minute, but it's on the same deprecation list, so anything built on it today has a little over three months left.

Build this in Wireflow
Make the same kind of asset on a canvas you control, with a free account and no credit card.
We ran one real file: a 14 minute speech
To check the per-run price we used a public domain recording, John F. Kennedy's 1961 inaugural address, which runs 14 minutes and 0 seconds. We sent it through the Speech to Text node with the task set to transcribe, the language set to English and word-level chunks, and then ran a 60 second cut of the same speech.
The full speech came back as about 1,390 words with a start and end time on every word, the last one ending at 833.6 seconds, after about 9 seconds of processing on the server. The 60 second cut finished in under 2 seconds. The Executions record showed 0 credits used for both runs, and the account balance was the same before and after.

Here is what the same two files cost on OpenAI, worked out from its per-minute rates.
| File | whisper-1 | gpt-transcribe | gpt-4o-mini-transcribe | Speech to Text node |
|---|---|---|---|---|
| 60 second cut | $0.006 | $0.0045 | about $0.003 | listed at 1 credit, our run logged 0 |
| 14 minute speech | $0.084 | $0.063 | about $0.042 | listed at 1 credit, our run logged 0 |
We'd still budget the listed 1 credit a run, because that's the price on the pricing page and the model page, even though the node's docs page says the node itself costs no credits. In both cases the charge didn't grow with the file, since the 14 minute speech and the 60 second cut were billed the same.
What an hour of audio costs, and when the per-minute rate wins
At 1.35 cents a credit, one listed run costs the same as about 2.3 minutes of whisper-1, 3 minutes of gpt-transcribe or 4.5 minutes of gpt-4o-mini-transcribe. Short clips under those lengths are cheaper on OpenAI's per-minute rates, while anything longer is cheaper as one flat run. How credits come out of a plan is set out in the credits and billing docs.
At OpenAI's rates, volume adds up like this:
- One hour of audio costs $0.36 on whisper-1, $0.27 on gpt-transcribe and about $0.18 on gpt-4o-mini-transcribe.
- 100 hours a month comes to $36 on whisper-1 or $27 on gpt-transcribe.
- At 1,000 hours a month gpt-transcribe bills $270, which is $90 less than whisper-1 for the same audio.
We only tested files up to 14 minutes on the node, so for an hour-long recording we'd suggest checking the credit estimate on the board before you run it.
Costs the per-minute price leaves out

OpenAI's transcription endpoint takes files up to 25 MB in flac, mp3, mp4, mpeg, mpga, m4a, ogg, wav or webm. An hour of speech saved as a 64 kbps mp3 is about 29 MB, so long recordings have to be compressed or split into chunks, and OpenAI's own guide warns that cutting mid-sentence can lose context and accuracy. Our 14 minute test file came to 6.7 MB at that bitrate, so it would go to OpenAI in a single request.
Word timestamps come from the timestamp_granularities[] parameter, which OpenAI's guide says works only on whisper-1. Captions, karaoke-style subtitles and cutting a video on a spoken phrase all need that word timing, so tools that rely on it will need a new source after 26 February 2027. The Speech to Text node returns word timestamps by default, the kind of timing that burned-in captions on a captioned short are built from.
Some other costs only show up once you're running at volume:
- OpenAI lists the Batch API as not supported for gpt-transcribe, so a large overnight backlog costs the same per minute as a single file.
- The Whisper route that turns speech in another language into English text,
/v1/audio/translations, is deprecated too. The Speech to Text node keeps a Translate to English task alongside Transcribe. - Speaker labels need gpt-4o-transcribe-diarize, which shuts down on the same date as whisper-1.
The open-source Whisper model is free to download under the MIT license, but self-hosting moves the bill to GPU or CPU time, servers and upkeep. For a team without spare GPUs, a hosted AI inference API usually costs less than keeping a machine running for occasional files.
Calling Speech to Text from code
To run the node from your own app, open the workflow below in your account and send an audio URL to the Run endpoint. A GET request on the same route lists the workflow's inputs and their node IDs, so you know which ID takes the file.
curl -X POST https://www.wireflow.ai/api/v1/workflows/YOUR_WORKFLOW_ID/run \
-H "Authorization: Bearer sk-your-api-key" \
-H "Content-Type: application/json" \
-d '{ "inputs": { "AUDIO_NODE_ID": "https://example.com/interview.mp3" } }'
The call returns 202 with an execution ID that you can poll, or you can add a webhook block so the transcript is posted to your server when it's ready, as the Webhooks docs show.
The transcript is a normal output on the board, so you can wire it into an LLM node for a summary, or into AI text to speech for a dubbed version, in the same run. Our guide to chaining multiple AI models in one API call walks through that pattern.
If you work in Claude, the MCP connector runs the same workflow from a chat without any code.
Try it yourself: Open the Speech to Text workflow. It holds the 60 second cut of the speech and its transcript, so you can swap in your own file and run it.
FAQ
How much does the OpenAI Whisper API cost?
whisper-1 costs $0.006 a minute of audio, which is $0.36 an hour. Its replacement, gpt-transcribe, costs $0.0045 a minute, or $0.27 an hour, on OpenAI's pricing page on 17 November 2026.
Is OpenAI shutting down whisper-1?
Yes. OpenAI deprecated whisper-1 on 26 August 2026 and shuts it down on 26 February 2027, together with gpt-4o-transcribe, gpt-4o-mini-transcribe and gpt-4o-transcribe-diarize. It points users to gpt-transcribe for recorded files and gpt-live-transcribe for live audio.
What is the cheapest OpenAI transcription model?
gpt-4o-mini-transcribe has the lowest rate at about $0.003 a minute, but it's deprecated and shuts down on 26 February 2027. Among the models that stay, gpt-transcribe is the cheapest at $0.0045 a minute.
How large a file can the Whisper API take?
OpenAI's transcription endpoint takes files up to 25 MB in flac, mp3, mp4, mpeg, mpga, m4a, ogg, wav or webm. Longer recordings need to be compressed or split into chunks under that size.
Which OpenAI model returns word-level timestamps?
OpenAI's guide says the timestamp_granularities[] parameter only works on whisper-1, so word timing on OpenAI depends on a model that shuts down on 26 February 2027. The Speech to Text node returns word timestamps by default.
How much does Speech to Text cost on Wireflow?
It's listed at 1 credit a run on the pricing page and the model page, whatever the file length. Our 60 second and 14 minute test runs on 17 November 2026 each logged 0 credits used.
Is Wireflow free?
You can open and change this workflow for free. A new account gets up to 60 starter credits and a free video, and each run uses credits. Plans start at $36 a month for 2,660 credits.
Conclusion
The OpenAI Whisper API costs $0.006 a minute until whisper-1 shuts down on 26 February 2027, and gpt-transcribe does the same job for $0.0045 a minute. That makes OpenAI the cheaper choice for a clip of a minute or two. On Wireflow the Speech to Text node is listed at 1 credit a run, whatever the length, and our 14 minute speech came back with word timestamps in about 9 seconds, ready to feed captions, a summary or a voiceover in the same workflow.

Co-Founder & Operations at Wireflow
Runs client operations and content strategy at Wireflow. Works directly with creative teams and agencies to build production AI workflows.
- Content Strategy
- Client Operations
Ask Claude for your next ad.
Connect once. Ask any time.
- No credit card
- You keep the rights to what you make
- Cancel anytime


