openai

/

whisper-large-v3

like 5.17k

Follow

OpenAI 28.1k

What I'm trying to do:

Build a lecture transcription app using Hugging Face's free Inference API with Whisper models for speech-to-text.
My setup:

• ✅ Valid Hugging Face API token (verified with /api/whoami-v2)
• ✅ Token has READ permissions
• ✅ Using Next.js API route to make server-side requests
• ✅ Sending audio as binary data (ArrayBuffer) with proper Content-Type header

request format

const response = await fetch(
"https://api-inference.huggingface.co/models/openai/whisper-base",
{
method: "POST",
headers: {
Authorization: Bearer ${apiKey},
"Content-Type": "audio/wav", // also tried audio/webm, audio/mpeg
},
body: audioBuffer, // ArrayBuffer from audio file
}
); “

Models tested (all return 410 Gone):

• ❌ openai/whisper-large-v3 → 410 Gone
• ❌ openai/whisper-large-v2 → 410 Gone
• ❌ openai/whisper-base → 410 Gone
• ❌ distil-whisper/distil-large-v2 → 410 Gone
• ❌ distil-whisper/distil-large-v3 → 410 Gone

Error received
HTTP 410 Gone
Response body: (HTML error page instead of JSON)

What I think is wrong:

• The models exist on Hugging Face Hub (I can see them, they have millions of downloads)
• My API key works (verified via API)
• BUT: HTTP 410 suggests these models aren't hosted on the free serverless Inference API

questions:

1. Are ANY Whisper/distil-whisper models available on the free api-inference.huggingface.co endpoint?
2. Do I need a paid Inference Endpoint to use Whisper models?
3. If not, which ASR (automatic speech recognition) models ARE available on the free tier?
4. Is there a way to check which models are available on the free Inference API before trying them?
   Additional context:

• Audio files are 20-25MB (within the 25MB limit)
• Testing with MP3, WEBM, and WAV formats
• Same 410 error regardless of audio format or model size
  Any guidance would be greatly appreciated! 🙏