Hi Groq team
— we’re seeing occasional HTTP 500 (Internal Server Error) responses from the audio transcription endpoint and wanted to report it with details.
Endpoint
POST https://api.groq.com/openai/v1/audio/transcriptions
Behavior
-
Requests sometimes fail with 500
-
The issue seems prompt-related (removing it works)
- but it also only happens to some audio files (very rarely) so a combination of both?
What we’ve observed
-
If we omit the
promptparameter, the request works consistently -
When a
promptis included, we sometimes get 500s -
Initially, removing
\nfrom the prompt appeared to fix it, but the issue has since returned even with simplified prompts -
The same original audio file works, but when we split it into left/right mono channels, one of the resulting files now fails intermittently
Example request (simplified)
curl --request POST \
--url https://api.groq.com/openai/v1/audio/transcriptions \
--header 'accept: application/vnd.api+json' \
--header 'authorization: Bearer <redacted>' \
--header 'content-type: multipart/form-data' \
--form file=@2.wav \
--form model=whisper-large-v3-turbo \
--form language=en \
--form 'prompt=Call about Company X. Transcribe exactly what is said; keep code-switching and diacritics.Do not add roles.' \
--form 'timestamp_granularities[]=word' \
--form response_format=verbose_json
Request IDs where this occurred
-
req_01kcsb3zg1e9vtnsr4drht0cav -
req_01kcsb4gvgf3mrmbsfm12ketym -
req_01kcsb8kt0e539ak06kbwze0zt -
req_01kcsba1rnen99hncj9pyxgxjr
Could you please check internal logs for these request IDs and let us know if there’s a known issue with prompts, newlines, or certain audio characteristics?
Happy to provide more samples or context if needed — thanks! ![]()