Error using openai/whisper-1 with OpenRouter provider in LiteLLM v1.83.10

[alihossein]

Check for existing issues

• I have searched the existing issues and checked that my issue is not a duplicate.

What happened?

I am trying to use openai/whisper-1 for audio transcription via the OpenRouter provider. Although the model is configured in the LiteLLM proxy, the request fails with a provider mapping error and a 400 Bad Request.

Questions & Context:

I have reviewed the documentation (https://docs.litellm.ai/docs/audio_transcription) but I'm still facing issues.

1. Is openai/whisper-1 currently unsupported when using the OpenRouter provider in LiteLLM?
2. If it's not supported, what is the recommended configuration or workaround to perform audio transcription through LiteLLM with OpenRouter?

Expected Behavior:
Successful audio transcription using Whisper-compatible models through OpenRouter.

Actual Behavior:
I receive a ValueError: Unmapped provider and an OpenAI BadRequestError indicating an invalid model name.

Steps to Reproduce

1. Add openai/whisper-1 to LiteLLM models via the Admin UI/config.
2. Set the provider to openrouter.
3. Make an audio transcription request (e.g., using the /v1/audio/transcriptions endpoint).
4. See the error in the logs/response.

import os
from pathlib import Path

import litellm

os.environ["OPENROUTER_API_KEY"] = "sk-xxxxxxxxx-Mr-A"
audio_file = Path("test.mp3")

result = litellm.transcription(
    api_key="sk-xxxxxxxxx-Mr-A",
    api_base="https://llm.xx.ir/v1",
    model="openai/whisper-1",
    audio=open("test.mp3", "rb"),
    language="fa",
    response_format="json",
    file=audio_file
)
print(result.text)

Relevant log output

openai.BadRequestError: Error code: 400 - {'error': {'message': "{'error': '/audio/transcriptions: Invalid model name passed in model=whisper-1. Call `/v1/models` to view available models for your key.'}", 'type': 'None', 'param': 'None', 'code': '400'}}

ValueError: Unmapped provider passed in. Unable to get the response.

What part of LiteLLM is this about?

Proxy

What LiteLLM version are you on ?

v1.83.10

Twitter / LinkedIn details

No response

[kalp9197]

Hey, I looked into this and the root cause is that OpenRouter doesn't support audio transcription endpoints — it only proxies chat/completion models. When LiteLLM routes openai/whisper-1 through OpenRouter, it strips the prefix and hits https://openrouter.ai/api/v1/audio/transcriptions, which OpenRouter simply doesn't handle — hence the Unmapped provider + 400.

Workaround for now:

Route Whisper directly to OpenAI instead of through OpenRouter:

result = litellm.transcription(
    model="whisper-1",
    file=open("test.mp3", "rb"),
    language="fa",
    response_format="json",
)

Or in proxy config:

model_list:
  - model_name: whisper-1
    litellm_params:
      model: whisper-1
      api_key: os.environ/OPENAI_API_KEY

Planning to raise a PR that adds an early validation check when OpenRouter is used for transcription — so instead of a confusing Unmapped provider error, users get a clear actionable message. Will also add a note in the audio transcription docs.

Will raise a PR soon!

[alihossein]

Hi,

Thanks for the clarification! To better understand the root cause, I did a quick experiment.

Interestingly, I successfully performed the transcription by making a direct requests call to OpenRouter using a Base64-encoded payload. It seems OpenRouter can handle whisper-1 if the data is sent as a JSON string rather than a binary file. Here is the working code:

import requests
import base64
import json

with open("test.mp3", "rb") as f:
    base64_audio = base64.b64encode(f.read()).decode("utf-8")

response = requests.post(
    url="https://openrouter.ai/api/v1/audio/transcriptions",
    headers={
        "Authorization": "Bearer YOUR_OPENROUTER_KEY",
        "Content-Type": "application/json",
    },
    data=json.dumps({
        "model": "openai/whisper-1",
        "input_audio": {
            "data": base64_audio,
            "format": "wav"
        }
    })
)
print(response.json()["text"])

However, when I try to replicate this behavior using litellm.completion (to bypass the binary file mapping issue in litellm.transcription), I get a BadRequestError:

# Error Log Summary:
# openai.BadRequestError: /chat/completions: Invalid model name passed in model=whisper-1.

It seems that when using litellm.completion, the request is routed to /chat/completions, but OpenRouter doesn't recognize whisper-1 as a valid chat model, even if input_audio is provided in the payload.

My Question:
Is there a way in LiteLLM to support this Base64/JSON approach for Whisper models through OpenRouter? This would be a great workaround for providers that don't support standard multipart/form-data audio endpoints but can process audio data within a JSON body.

What are your thoughts on this?

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.