OpenAI Whisper API pricing: OpenAI docs, September 2026
Whisper costs $0.006 a minute ($0.36 an hour). OpenAI lists gpt-transcribe at $0.0045 and gpt-4o-mini-transcribe at an estimated $0.003. These prices were checked 28 September 2026; accuracy needs a separate test.
OpenAI’s Whisper API costs $0.006 per minute of audio, which is $0.36 an hour, on OpenAI’s pricing pages checked 28 September 2026.1 Whisper is not OpenAI’s cheapest transcription model. gpt-4o-mini-transcribe ranks first at an estimated $0.003 per minute, and gpt-transcribe costs $0.0045.1 OpenAI prints these figures with a dollar sign and states neither the currency code nor the tax treatment.1
TL;DR
- Whisper’s $0.006 a minute ties with gpt-4o-transcribe and gpt-4o-transcribe-diarize for third place among eight transcription models.1
- For recorded audio, gpt-transcribe costs 25 per cent less than Whisper and gpt-4o-mini-transcribe costs half as much. At 1,000 hours a month that is $270 or $180 against Whisper’s $360.1
- Live transcription costs nearly three times as much as Whisper: gpt-live-transcribe and gpt-realtime-whisper are both $0.017 a minute. Live translation costs $0.034.1
- Text-to-speech is priced per character or per token, never per minute: tts-1 at $15.00 and tts-1-hd at $30.00 per 1M characters, and gpt-4o-mini-tts at $0.60 per 1M text input tokens and $12.00 per 1M audio output tokens.1
- OpenAI’s pricing pages publish no accuracy figures. Price decides the shortlist; a test on your own audio decides the model. This price table sits in our model infrastructure research.
How this ranking was built
Glamdring Research ranked every model in the Transcription models table of OpenAI’s API pricing documentation by one field: Estimated cost, per minute of audio.1 Prices appear per minute or per million tokens as published, from OpenAI’s pricing pages, checked 28 September 2026. The documentation was captured with every “All models” list expanded. The default view of the same table shows six models. Whisper and gpt-4o-transcribe-diarize appear only in the expanded list.2,1
The table holds eight models and all eight are ranked. Equal prices share a rank. Three models tie at $0.006 and two at $0.017.1 We checked each figure against OpenAI’s main API pricing page, captured the same day. That page lists gpt-transcribe at “$0.0045 per minute / $0.00008 per second” and gpt-live-transcribe at “$0.017 per minute / $0.00028 per second”. Both match the documentation.3 The main page does not list Whisper.3
Hourly and monthly figures are our arithmetic: the per-minute price times 60, then times the hours stated. Text-to-speech and voice-conversation models are listed separately and left unranked, because OpenAI prices them per character, per token or per session minute.1 Our research methodology sets out how company-published prices are treated: they are OpenAI’s statements, not measured market prices.
How much does each OpenAI transcription model cost per minute?
Whisper costs $0.006 per minute on OpenAI’s API, the same as gpt-4o-transcribe and gpt-4o-transcribe-diarize.1 gpt-4o-mini-transcribe is the cheapest at an estimated $0.003 per minute, followed by gpt-transcribe at $0.0045.1 The live models, gpt-live-transcribe and gpt-realtime-whisper, cost $0.017, and gpt-realtime-translate costs $0.034.1 These are OpenAI’s published prices, checked 28 September 2026.
| Rank | Model | OpenAI’s use case | Estimated cost ($ per minute, as printed) | Per hour (our arithmetic) | Token rates shown (per 1M, input / output) |
|---|---|---|---|---|---|
| 1 | gpt-4o-mini-transcribe | Transcription | $0.003 | $0.18 | $1.25 / $5.00 |
| 2 | gpt-transcribe | Transcription | $0.0045 | $0.27 | None shown |
| 3 | gpt-4o-transcribe | Transcription | $0.006 | $0.36 | $2.50 / $10.00 |
| 3 | gpt-4o-transcribe-diarize | Transcription + diarization | $0.006 | $0.36 | $2.50 / $10.00 |
| 3 | Whisper | Transcription | $0.006 | $0.36 | None shown |
| 6 | gpt-live-transcribe | Live transcription | $0.017 | $1.02 | None shown |
| 6 | gpt-realtime-whisper | Live transcription | $0.017 | $1.02 | None shown |
| 8 | gpt-realtime-translate | Live translation | $0.034 | $2.04 | None shown |
Model names, use cases and prices are OpenAI’s, as the documentation gives them.1
The table mixes two billing shapes. Three models carry per-token rates beside the per-minute figure: gpt-4o-transcribe and gpt-4o-transcribe-diarize at $2.50 input and $10.00 output per 1M tokens, and gpt-4o-mini-transcribe at $1.25 and $5.00.1 For those three, the per-minute figure is OpenAI’s estimate of what the tokens cost. The other five show no token rate. OpenAI’s main pricing page prints gpt-transcribe’s figure as its price, with a per-second equivalent.3
OpenRouter gives Whisper’s API model ID as whisper-1. It routes requests for the model to OpenAI, lists it at “$0.0001/second” and describes it as “Priced per minute of audio duration, billed to the nearest second.”4 That per-second rate is $0.006 divided by 60, so the two sources agree.
At volume, the gaps widen in absolute terms. 1,000 hours of recorded audio a month costs $360 on Whisper, $270 on gpt-transcribe and $180 on gpt-4o-mini-transcribe at the listed rates. The same 1,000 hours through gpt-live-transcribe costs $1,020.1
Is Whisper the cheapest way to transcribe audio on OpenAI’s API?
No. Two OpenAI transcription models cost less than Whisper’s $0.006 per minute: gpt-transcribe at $0.0045 and gpt-4o-mini-transcribe at an estimated $0.003.1 OpenAI describes gpt-transcribe as “A speech-to-text model for asynchronous and batch workloads”, the recorded-audio work that Whisper’s “Transcription” row also covers.3,1 Glamdring Research’s verdict on price alone: a new batch workload has no reason to start on Whisper.
The decision changes when the transcript needs more than text. gpt-4o-transcribe-diarize, listed for “Transcription + diarization”, costs the same $0.006 per minute as Whisper.1 Speaker labels therefore carry no per-minute premium over Whisper on OpenAI’s list. The decision also changes for live audio. gpt-realtime-whisper, the live-transcription row that carries Whisper’s name, costs $0.017 per minute. So does gpt-live-transcribe, which OpenAI describes as transcribing “speech live as the speaker talks”.1,3 Streaming costs about 2.8 times the recorded-file rate.
The constraint is accuracy, and OpenAI’s pricing pages publish none. The cheapest model wins only if its transcripts are good enough for the work. gpt-4o-mini-transcribe’s $0.003 is also an estimate built on token rates, so its billed cost moves with the tokens a file produces.1 Run one representative hour of your own audio through gpt-transcribe, gpt-4o-mini-transcribe and Whisper. Compare the word errors and the billed amount in your usage dashboard, which OpenAI says shows usage “during the current and past billing cycles”.3
What do OpenAI’s text-to-speech models cost?
OpenAI prices text-to-speech per character or per token, never per minute of audio produced. tts-1 costs $15.00 per 1M characters and tts-1-hd costs $30.00 per 1M characters. gpt-4o-mini-tts costs $0.60 per 1M text input tokens and $12.00 per 1M audio output tokens.1 All three sit in the Realtime and audio generation models table of OpenAI’s pricing documentation, checked 28 September 2026.
| Model | Input price | Output price | Unit |
|---|---|---|---|
| tts-1 | $15.00 | None listed | per 1M characters of text |
| tts-1-hd | $30.00 | None listed | per 1M characters of text |
| gpt-4o-mini-tts | $0.60 (text) | $12.00 (audio) | per 1M tokens |
| gpt-live-1 | $0.05 per minute | Backend model and tool usage charged separately | per session minute |
| gpt-realtime-2.1 | $32.00 (audio) | $64.00 (audio) | per 1M tokens |
| gpt-realtime-2.1-mini | $10.00 (audio) | $20.00 (audio) | per 1M tokens |
| gpt-audio-1.5 | $32.00 (audio) | $64.00 (audio) | per 1M tokens |
| gpt-audio-mini | $10.00 (audio) | $20.00 (audio) | per 1M tokens |
Prices from OpenAI’s pricing documentation.1 The realtime models also carry text and image rates; gpt-realtime-2.1, for example, charges $4.00 per 1M text input tokens and $24.00 per 1M text output tokens.1
The character-priced models are the simplest to budget. A 50,000-character script costs $0.75 on tts-1 and $1.50 on tts-1-hd. 1,000 characters cost $0.015 and $0.03. gpt-4o-mini-tts cannot be converted the same way from the captured pages, because they don’t state how many audio tokens a minute of speech produces.
gpt-live-1 is the only voice model priced per minute. OpenAI says its sessions “are billed per second, without rounding up to a whole minute”, and that “Backend model and tool usage is charged separately.”1 At $0.05 a minute, an hour of conversation costs $3.00 before those backend charges.
How does OpenAI define the per-minute figure?
OpenAI labels the ranked column “Estimated cost” and heads the table “Prices per 1M tokens unless noted.”1 For models with token rates, the per-minute figure is OpenAI’s estimate of the token cost of a minute of audio. For Whisper, gpt-transcribe and the live models, no token rate appears, so the per-minute figure is the only price shown.1
The distinction matters for budgeting. A per-minute price scales with audio length alone. A token price scales with what the model reads and writes, so two files of equal length can bill differently. Our explainer on what controls AI inference cost covers why token counts vary with the work.
Two adjustments sit outside the table. Regional processing endpoints for data residency carry “a 10% uplift for models released on or after March 5, 2026, that are eligible for data residency.”1 FedRAMP endpoints carry a 10 per cent uplift over standard rates.1 The captured pages don’t say which speech models qualify for data residency or when each was released.
OpenAI’s main pricing page shows a mode selector above its voice models reading “Standard”, “Batch -50%” and “Data residency +10%”.3 The capture records standard prices only. Whether the batch discount applies to each speech model was not confirmed.
Why do other published Whisper prices disagree?
The third-party sources captured for this research price Whisper at $0.006 per minute or its hourly or per-second equivalent, matching OpenAI. They differ from OpenAI on what else a transcript costs and on which models exist.4,5,6
BrassTranscripts, a transcription service, lists three OpenAI options: Whisper and GPT-4o Transcribe at $0.006 and GPT-4o Mini Transcribe at $0.003.6 It omits gpt-transcribe at $0.0045, which OpenAI’s current pages list.1 It also adds “separate diarization (≈$0.005/min)” on top of Whisper, stating that “Whisper API does NOT identify speakers.”6 OpenAI’s own list includes gpt-4o-transcribe-diarize at $0.006 per minute, so diarization on OpenAI’s API need not add a second service.1
Billed amounts can also differ from list arithmetic. In November 2023 a developer on OpenAI’s community forum reported transcribing 2,333,349 seconds of audio, estimating $233.34 and being charged $397.08. The poster concluded the “real cost should be 0.010 $ per minute”.7 The captured thread doesn’t show the replies’ content, so the cause is unresolved here. That is one user’s report from 2023, not a current OpenAI rate.
The Whisper rate itself has held. A December 2023 forum post put the Whisper API at “$0,36/h”, the same $0.006 a minute OpenAI lists in September 2026.5,1 The field around it differs. OpenAI’s September 2026 table lists two recorded-audio models below Whisper’s price and shows Whisper only in its expanded view.1,2
What the ranking cannot tell you
The ranking orders OpenAI’s prices. It cannot show which model transcribes your audio most accurately, because OpenAI’s pricing pages publish no accuracy or error-rate figures. A cheaper model that needs more correction can cost more per usable transcript.
Three figures are estimates built on token rates: gpt-4o-mini-transcribe, gpt-4o-transcribe and gpt-4o-transcribe-diarize.1 Their billed cost can move away from the listed minute rate. The prices are OpenAI’s statements, printed with a dollar sign and no currency code or tax treatment.1 Local-currency prices were not captured.
The pricing pages carry no deprecation note for Whisper. We did not check OpenAI’s separate deprecation schedule. OpenAI’s main pricing page lists GPT-Live-Transcribe, GPT-Transcribe and GPT-Realtime-Translate as its speech-to-text and translation models and leaves Whisper out.3 Teams weighing transcription outside OpenAI can start from our survey of alternatives to the OpenAI API.
Frequently asked questions
Is the OpenAI Whisper API free to use?
No free allowance for Whisper appears on OpenAI’s captured pricing pages; the documentation lists it at $0.006 per minute.1 OpenAI’s pricing FAQ also states that its APIs “are billed separately from ChatGPT Plus, Business, Enterprise and Edu”, so a ChatGPT subscription does not cover API transcription.3
Is OpenAI Whisper free and open source?
The Whisper model is open source; OpenRouter describes it as “OpenAI’s open-source automatic speech recognition model”.4 Running it on your own hardware avoids OpenAI’s per-minute charge, but you pay for the compute and its upkeep. In December 2023 one forum user compared the API’s “$0,36/h” with “$0,39/h” for a rented 4090 spot instance on RunPod, with both taking “roughly 2 minutes for an hour of speech”.5 Those are one user’s 2023 figures, not current rates.
How much do 1,000 tokens cost on OpenAI’s speech models?
On gpt-4o-transcribe, 1,000 input tokens cost $0.0025 and 1,000 output tokens cost $0.01, from its $2.50 and $10.00 per 1M rates.1 On gpt-4o-mini-transcribe the figures are $0.00125 and $0.005. On gpt-4o-mini-tts, 1,000 text input tokens cost $0.0006 and 1,000 audio output tokens cost $0.012.1 Whisper has no token price; it bills by audio length.
How much does the OpenAI API cost outside speech?
OpenAI’s flagship text models are priced per 1M tokens. At standard rates for short context, GPT-6 Luna costs $0.10 input and $0.50 output, GPT-6 Sol $2.00 and $10.00, and GPT-6 Astra $10.00 and $50.00.3 The Batch API takes 50 per cent off inputs and outputs for tasks run asynchronously “over 24 hours”.3
Does Whisper charge for a full minute on a short file?
OpenAI’s captured pricing pages don’t state Whisper’s rounding rule. OpenRouter, which routes whisper-1 requests to OpenAI, says it is “Priced per minute of audio duration, billed to the nearest second” at $0.0001 per second.4 On those terms, a 30-second clip costs $0.003.
Sources checked
Company-owned pages establish what a company says. They do not prove a market conclusion. Each source is dated so readers can judge each claim.
