413 Request Entity Too Large or InvalidRequestError
openai.InvalidRequestError (HTTP 413 or API validation error)
Stack trace
openai.APIStatusError: Error code: 413 -- {"error": {"message": "Request body too large", "type": "invalid_request_error", "param": null, "code": "request_too_large"}} or openai.InvalidRequestError: Error code: 400 -- {"error": {"message": "file size must be below 25 MB", "type": "invalid_request_error"}} Why it happens
OpenAI's Whisper API (whisper-1 model) has a hard 25MB file size limit to control infrastructure costs and ensure reasonable latency. When you attempt to transcribe an audio file larger than this limit via client.audio.transcriptions.create(), the API rejects the request before processing. This is a documented API constraint, not a bug.
Detection
Check file size before uploading: if os.path.getsize(audio_file_path) > 25 * 1024 * 1024, you need compression or streaming. Add logging to track file sizes in production: logger.info(f'Audio file size: {os.path.getsize(path) / 1024 / 1024:.2f}MB').
Causes & fixes
Audio file is genuinely larger than 25MB without compression (e.g., WAV, uncompressed MP3, or long recording)
Compress the audio file using ffmpeg before upload: ffmpeg -i input.wav -acodec libmp3lame -ab 64k output.mp3 reduces file size to ~30-40% with minimal quality loss at 64 kbps bitrate
Uploading a lossless or high-bitrate audio file (FLAC, WAV at 320kbps, or 48kHz PCM)
Transcode to MP3 or M4A at 64-128 kbps bitrate using ffmpeg: ffmpeg -i input.flac -acodec libmp3lame -ab 128k output.mp3, which reduces 25MB+ files to <5MB
Audio file is split into multiple segments and each segment exceeds 25MB individually
Pre-split the audio into 20-minute chunks using pydub or ffmpeg before calling Whisper: from pydub import AudioSegment; chunks = [audio[i:i+20*60*1000] for i in range(0, len(audio), 20*60*1000)], then loop through client.audio.transcriptions.create() for each chunk
OpenAI API library is buffering the entire file in memory before sending, inflating apparent size
Use open(file_path, 'rb') context manager to stream the file directly: with open('audio.mp3', 'rb') as f: response = client.audio.transcriptions.create(model='whisper-1', file=f), which avoids memory buffering
Code: broken vs fixed
import os
from openai import OpenAI
client = OpenAI(api_key=os.environ.get('OPENAI_API_KEY'))
# This will fail if audio.wav > 25MB
response = client.audio.transcriptions.create(
model='whisper-1',
file=open('audio.wav', 'rb') # ❌ NO SIZE CHECK — if >25MB, 413 error
)
print(response.text)import os
import subprocess
from pathlib import Path
from openai import OpenAI, APIStatusError
client = OpenAI(api_key=os.environ.get('OPENAI_API_KEY'))
def compress_audio_if_needed(audio_path: str, max_size_mb: int = 25) -> str:
"""Compress audio to 128kbps MP3 if file exceeds max_size_mb."""
file_size_mb = os.path.getsize(audio_path) / (1024 * 1024)
if file_size_mb > max_size_mb:
output_path = Path(audio_path).stem + '_compressed.mp3'
# ✅ FIXED: Compress to 128kbps MP3 to reduce file size
subprocess.run([
'ffmpeg', '-i', audio_path, '-acodec', 'libmp3lame',
'-ab', '128k', '-y', output_path
], check=True, capture_output=True)
new_size_mb = os.path.getsize(output_path) / (1024 * 1024)
print(f'Compressed {file_size_mb:.2f}MB → {new_size_mb:.2f}MB')
return output_path
return audio_path
try:
audio_file = compress_audio_if_needed('audio.wav')
with open(audio_file, 'rb') as f:
response = client.audio.transcriptions.create(
model='whisper-1',
file=f
)
print(f'Transcription: {response.text}')
except APIStatusError as e:
if e.status_code == 413:
print(f'Error: File still too large. Try increasing compression bitrate.')
raiseWorkaround
If you cannot install ffmpeg in your environment, use pydub to split audio into 20-minute segments, upload each segment separately, and concatenate transcription results: from pydub import AudioSegment; audio = AudioSegment.from_file('large.wav'); chunks = [audio[i:i+20*60*1000] for i in range(0, len(audio), 20*60*1000)]; transcripts = []; [transcripts.append(client.audio.transcriptions.create(model='whisper-1', file=chunk.export(format='mp3')).text) for chunk in chunks]; final_text = ' '.join(transcripts).
Prevention
Always check file size before calling Whisper API (if os.path.getsize(file) > 25MB, compress). For production pipelines, implement automatic compression middleware that targets 64-128 kbps MP3 as the default upload format. For very large audio archives, pre-process audio at ingest time, not at transcription request time.