Tool profile
OpenAI speech-to-text API (gpt-4o-transcribe): cost, fit and limits
The OpenAI speech to text API is a set of audio transcription models offered through the OpenAI platform, with gpt-4o-transcribe named as the speech to text model on its documentation. It is billed per token of audio input, not per minute.
What to know first
OpenAI's gpt-4o-transcribe model is priced at $2.5 per 1M audio input tokens on the model page, and the pricing pages opened give no per-minute or per-hour figure. No free tier is stated. The model is documented on the OpenAI developer site, and the opened pages say nothing on retention, training or hosting region. The legal entity page returned an error when opened.
OpenAI speech-to-text API (gpt-4o-transcribe) at a glance
| Maker | OpenAI (legal entity not stated on the opened pages), Not stated on the vendor pages |
|---|---|
| Free plan | No |
| Entry price | $2.5 per 1M audio input tokens |
| API access | Yes |
| Category | Best AI Speech to Text APIs in 2026, rank 7 of 12 |
| Website | openai.com/api/pricing/ |
Who OpenAI speech-to-text API (gpt-4o-transcribe) is best for
Teams already building on OpenAI models
Transcription lives on the same platform account and key as other models, which shortens integration for products already calling the API.
Prototypes that feed text to a language model
Audio converted to text can go straight into the same vendor's text models, keeping a voice feature in one billing relationship.
Engineers comfortable with token accounting
The price is per token, so teams that already monitor token usage for text models can reuse the same tracking and budgeting habits.
OpenAI speech-to-text API (gpt-4o-transcribe) plans and pricing
| Plan | Price | What it includes |
|---|---|---|
| API, pay per token | $2.5 per 1M audio input tokens (gpt-4o-transcribe) | Audio transcription through the model, billed on tokens used. No separate speech plan or monthly fee is listed on the opened pages. |
| Free tier | Not stated on the opened pages | Neither the pricing page nor the model page mentions a trial credit or free monthly allowance for transcription. |
| Per-minute rate | Not published on the model page | Cost per hour of audio must be estimated by measuring how many tokens a sample recording consumes. |
Checked November 2026.
What OpenAI speech-to-text API (gpt-4o-transcribe) does
- gpt-4o-transcribe model
- The model page names gpt-4o-transcribe as OpenAI's speech to text model and lists audio input as the billed quantity. Other audio models from the vendor are not covered on the opened pages, so only this one is described.
- Token-based audio pricing
- Audio input is priced at $2.5 per 1M tokens, so the cost of an hour depends on how many tokens that audio produces. A buyer who wants a per-hour number must measure tokens on a sample, because the page does not convert.
- Developer documentation
- The model is documented on OpenAI's developer docs site, which is where parameters, formats and supported languages would be confirmed. Supported file formats and size limits were not captured, so those details live in the documentation, not here.
- No standalone speech plan
- The opened pages list no subscription for speech, so transcription spend is part of general API usage on the account. Cost control therefore depends on the account's usage limits and budgets, not on a speech-specific plan.
How OpenAI speech-to-text API (gpt-4o-transcribe) handles your data
The model page for gpt-4o-transcribe does not state a retention period, a training policy, a hosting region or certifications, so none are claimed in this entry. OpenAI publishes data usage and enterprise privacy documentation separately, and those pages were not opened. The OpenAI terms page returned a 403 error when we tried to open it, so the legal entity is also unconfirmed. Review OpenAI's current API data policies and business terms before sending personal recordings.
Where OpenAI speech-to-text API (gpt-4o-transcribe) falls short in practice
- No per-minute price. The model page gives only a per-token figure, so a buyer must run sample audio and read the token count to estimate cost per hour.
- No stated free tier. Neither opened page mentions a trial credit or free allowance for transcription, so evaluation starts with a billed request. A small test therefore costs real money, however little, unless the account has credit from elsewhere.
- Policies not on the model page. Retention, training and region are not stated there, which sends compliance teams to separate OpenAI documents we did not read. Teams in regulated industries should plan extra time to collect those documents before a pilot.
Works with
Alternatives to OpenAI speech-to-text API (gpt-4o-transcribe)
OpenAI speech-to-text API (gpt-4o-transcribe): common questions
How much does the OpenAI transcription API cost?
The gpt-4o-transcribe model page lists audio input at $2.5 per 1M tokens. It gives no per-minute figure, so the cost per hour depends on how many tokens your audio produces. Send a ten minute sample, read the token count from the response, and scale it up.
Does OpenAI offer a free tier for transcription?
No free tier is stated on the opened pricing and model pages. A buyer should assume evaluation is billed at the token rate, and should check the account dashboard for any credit that applies to a new organization, since that detail was not on the pages we read.
Is the OpenAI transcription model billed per minute?
No. The pages opened bill audio by tokens, and no per-minute price appears on the model page. That differs from vendors such as Amazon, which print per-minute rates, so any comparison needs a conversion based on a measured sample of your own audio.
Does OpenAI keep or train on audio sent to the API?
The model page does not state a retention or training policy, so this site makes no claim. OpenAI's separate data usage and business privacy pages cover it and were not opened for this entry. Read them, and confirm in writing, before sending regulated recordings.
Why is the OpenAI legal entity not listed here?
The OpenAI terms page returned a 403 error when we tried to open it, and none of the pages that did open named the contracting entity. This entry therefore records the maker as OpenAI only, with the entity unconfirmed. Check the current terms during contracting.