Whisper large-v3 API

OpenAI, open modelwhisper-large-v3Model source: Hugging Face

Turns audio into text: meetings, calls, interviews and videos.

Turn recordings, calls, meetings and videos into text. Whisper large-v3 is OpenAI's open-source transcription model, in its most accurate version. On GPU Flow it costs €0.0015 per minute of audio, billed by the second, and uses the same endpoint as the OpenAI audio API.

What it does

Speech to text

stt

How it is charged

Each minute of audio

0.0015 €

Where it runs

Spain (EU)

MIG B200

Brain size

Small and fast

Transcription

What it is good at

Meetings, calls and interviews. Several speakers, background noise, accents. Large-v3 is the most robust version of Whisper.
Timestamps for subtitles. With verbose_json you get segments with start and end times, and word-level times if you ask for them: the basis for building your own subtitles.
Spanish and co-official languages. Spanish, Catalan, Galician and Basque among the 99 languages, with automatic detection or a forced language.

It transcribes; it does not translate, summarise or identify who is speaking. For those, send the transcript to a language model on the same API.

Pricing

€ per minute of audio, VAT included

Each minute of audio (use)0.0015 €

try it with 5 € freeActivated with your card, and you only pay for what you use.

For your technical team

Upload the audio file. With response_format=verbose_json it also returns the duration.

curl https://api.gpuflow.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $GPUFLOW_API_KEY" \
  -F model=whisper-large-v3 \
  -F [email protected] \
  -F response_format=verbose_json

Limits and formats

Requests per minute: 1,000 per key, with bursts up to double; raisable to 5,000 from the dashboard.
Input audio: Whatever FFmpeg decodes: mp3, m4a, wav, ogg/opus, webm, flac and more.
Response: json, text or verbose_json, with segments and, if you ask, word timings.
Not available: srt or vtt subtitles, translation and streaming.

Frequently asked questions

How much does Whisper large-v3 cost?

€0.0015 per minute of audio, VAT included. No monthly fee.

Where does it run?

On our NVIDIA B200 GPUs, in a data centre in Spain.

Is the Whisper API compatible with OpenAI's?

Yes: same /v1/audio/transcriptions endpoint and parameters; change the base URL and the key. It responds in json, text or verbose_json; srt and vtt output, the translation endpoint and streaming are not available.

Which languages and audio formats does it support?

99 languages, with automatic detection or one you set with language. It accepts any audio FFmpeg can decode, such as mp3, m4a, wav, ogg/opus, webm or flac.

Does it identify speakers?

No. It transcribes what is said; speaker identification is not included.

Is the audio stored?

No. The audio is processed in memory and the transcript comes back in the same call; we do not keep it afterwards and it is not used for training.