MAI-Transcribe 2

MAI-Transcribe 2 (microsoft/mai-transcribe-2) is Microsoft’s speech recognition model: 60 languages, Russian included, and one of the strongest results on the multilingual FLEURS benchmark. It is billed by audio duration rather than tokens, so the cost of a transcription is known up front, from the length of the file.

Parameters

POST /v1/audio/transcriptions, multipart/form-data:

Field	Description
file	Audio file, required
model	microsoft/mai-transcribe-2
language	Language code (ru, en, …); omitted, the language is detected automatically
response_format	json by default

Example

Python

from openai import OpenAI

client = OpenAI(base_url="https://api.mixen.ai/v1", api_key=MIXEN_API_KEY)

with open("meeting.mp3", "rb") as f:
    result = client.audio.transcriptions.create(
        model="microsoft/mai-transcribe-2",
        file=f,
        language="en",
    )
print(result.text)

cURL

curl https://api.mixen.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $MIXEN_API_KEY" \
  -F file=@meeting.mp3 \
  -F model=microsoft/mai-transcribe-2 \
  -F language=en

Things to keep in mind

• Pass language when you know it. Auto-detection slips on short and noisy recordings, and a wrong language ruins the whole transcript.
• The model is chosen by the model field. whisper-1 is accepted for OpenAI client compatibility and routes to the default model — for MAI-Transcribe 2 pass the full key.
• There are cheaper and pricier options. The full recognition line-up with prices is in the catalog; the endpoint itself is covered in Speech & embeddings.