Ink 2.0

Real-time speech-to-text model, built for the lowest word error rate and native turn detection in streaming voice agents.

Talk to an engineer

Model details

Developed by
Cartesia
Use case
transcription

Example usage

Ink 2 is Cartesia's fastest and most accurate streaming speech-to-text model, ranked #1 for lowest word error rate and best turn detection of any streaming STT. It transcribes structured data like phone numbers, dates, and emails right the first time, and knows when a speaker starts and finishes with no separate VAD required.

See full documentation

Input

1import requests
2
3url = "https://api.cartesia.ai/stt"
4
5files = { "file": ("example-file", open("example-file", "rb")) }
6payload = {
7    "model": "ink-whisper",
8    "language": "en",
9    "timestamp_granularities[]": "word"
10}
11headers = {
12    "Cartesia-Version": "2026-03-01",
13    "Authorization": "$CARTESIA_API_KEY"
14}
15
16response = requests.post(url, data=payload, files=files, headers=headers)
17
18print(response.text)

JSON output

1{
2    "type": "transcript",
3    "text": "<string>",
4    "request_id": "<string>",
5    "is_final": true,
6    "language": "<string>",
7    "duration": 123,
8    "words": [
9        {
10            "word": "<string>",
11            "start": 123,
12            "end": 123
13        }
14    ]
15}