Transcribe audio

Deepgram

Speech-to-text API that transcribes recorded or live audio into text with word timings and optional speaker labels.

Rate limits

Concurrent pre-recorded requests: 50 per project for Nova-3, Nova-2, Nova, Enhanced and Base (Pay as you go and Growth)2
Concurrent pre-recorded requests (other page): 100 per project for Nova, Base and Enhanced3
Concurrent streaming connections: 150 on Pay as you go, 225 on Growth (North America); Enterprise set to peak traffic1
Concurrent requests with diarization: 50 pre-recorded and 50 streaming in North America; 25 each in Europe, Australia and India2
Concurrent Whisper Cloud requests: 3 pre-recorded; not available outside North America2
Scope of limits: Per project, not per account or API key; extra projects on self-serve accounts get 1 concurrent stream2

Other limits

Pre-recorded file size: 2 GB3
Pre-recorded processing time: Requests over 10 minutes (Nova, Base, Enhanced) or 20 minutes (Whisper) return 504 Gateway Timeout3
Temporary token lifetime: 30 seconds13
Log retention: Request metadata and usage logs retrievable for 90 days6

Read from Deepgram’s own pages on 2026-11-04. Every number links to the page a value came from. For agents, the same facts as markdown: /tools/transcribe-audio/deepgram.md