Skip to main content
The /v1/audio/transcriptions endpoint transcribes audio files using Whisper Large v3 — optimized by the Boole compiler for fast inference. Submit an audio file and receive a text transcript. The endpoint accepts a wide range of audio formats and supports automatic language detection, custom prompts for improved accuracy, and multiple output formats including plain text, SRT, and VTT.

​
Endpoint

​
Request

Send requests as multipart/form-data.
​
file
required
The audio file to transcribe. Supported formats: mp3, mp4, mpeg, mpga, m4a, wav, webm. Maximum file size: 25 MB.
​
string
required
Must be whisper-large-v3.
​
string
ISO-639-1 language code of the spoken language, e.g. en, fr, de. When omitted, the language is detected automatically from the audio.
​
string
Optional text to guide the model’s style or supply context, such as proper nouns or domain-specific vocabulary. Does not need to match the spoken language.
​
string
Format of the transcript output. One of: json (default), text, srt, vtt, or verbose_json. verbose_json includes segment-level timestamps and metadata.
​
number
Sampling temperature between 0 and 1. At 0 (default), the model uses greedy decoding for the most deterministic output.

​
Example Request

​
Example Response

​
Pricing

Whisper Large v3 is billed at $0.02 per minute of audio, rounded up to the nearest second. See Pricing for the full rate card and details on how audio duration is measured.
Pass a language code when you know the spoken language — it improves accuracy and reduces latency.