gpt-4o-transcribe client.audio.transcriptions.create fails on all supported audio types (“This model does not support the format you provided.”)

Confirm this is an issue with the Python library and not an underlying OpenAI API

• This is an issue with the Python library

Describe the bug

When calling client.audio.transcriptions.create(model="gpt-4o-transcribe", file=...) I consistently receive:

openai.BadRequestError: Error code: 400 - {'error': {'message': 'This model does not support the format you provided.', 'type': 'invalid_request_error', 'param': 'messages', 'code': 'unsupported_format'}}

regardless of supplying the file as:

• wav (PCM‑16, 16 kHz, mono)
• mp3
• m4a (AAC)
• an in‑memory BytesIO buffer with .name set to the proper extension

The same code works fine with the whisper-1 model, but gpt-4o-transcribe/mini always rejects the input, even though I have verified via ffprobe/sox that my files are valid PCM‑16 WAV, standard MP3, or AAC.

To Reproduce

from openai import OpenAI
from dotenv import load_dotenv
from io import BytesIO
import os

load_dotenv()
client = OpenAI()

# 1) Try WAV on disk
with open("audio.wav", "rb") as f:
    resp = client.audio.transcriptions.create(
        model="gpt-4o-transcribe",
        file=f
    )
print(resp)

# 2) Try MP3 on disk
with open("audio.mp3", "rb") as f:
    resp = client.audio.transcriptions.create(
        model="gpt-4o-transcribe",
        file=f
    )
print(resp)

# 3) Try M4A on disk
with open("audio.m4a", "rb") as f:
    resp = client.audio.transcriptions.create(
        model="gpt-4o-transcribe",
        file=f
    )
print(resp)

# 4) Try BytesIO wrapper
data = open("audio.wav", "rb").read()
buffer = BytesIO(data)
buffer.name = "audio.wav"
buffer.seek(0)
resp = client.audio.transcriptions.create(
    model="gpt-4o-transcribe",
    file=buffer
)
print(resp)

Code snippets

from openai import OpenAI

from dotenv import load_dotenv

load_dotenv()

client = OpenAI()

audio_file = open("temp/audio.wav", "rb")
transcription = client.audio.transcriptions.create(
    model="gpt-4o-transcribe", file=audio_file
)
print(transcription.text)

OS

Windows 11

Python version

3.13

Library version

openai==1.97.0

Start by reproducing the reported calls through client.audio.transcriptions.create using the WAV, MP3, M4A, and BytesIO examples in the issue. Inspect the transcription request path and compare behavior with whisper-1; done means the supported audio inputs no longer fail for gpt-4o-transcribe, or the library/API boundary is clearly identified.