Deepgram

Speech API for pre-recorded and streaming speech-to-text, text-to-speech and a voice agent API. SDKs for Python, JavaScript, Go, .NET and Java.

Deepgram at a glance

From official pages, checked Nov 15, 2026. Tap the arrow for each source.

Free tier: Trial credits
$200 in free credit on Pay As You Go, no credit card required.
Starting price: $0.0043 per minute
Nova-3 pre-recorded, monolingual, Pay As You Go.
Rate limit: 50 concurrent requests
Pay As You Go, speech-to-text REST API (150 for streaming WSS).
Size limit: 2 GB per file
Maximum file size for pre-recorded audio.
Official SDKs: PythonJavaScript / TypeScript
Trains on your data: Opt-out
Requests join the Model Improvement Program unless mip_opt_out=true is set.
EU hosting: Yes
EU endpoint api.eu.deepgram.com keeps processing in the EU.
Compliance: SOC 2GDPRHIPAA

Send an audio URL to Deepgram's pre-recorded API and get a smart-formatted Nova-3 transcript back from Python.

Nov 11, 2026
Deepgram Self-Hosted 261110: unsupported language hints now fail with HTTP 400 (November 10, 2026)
Self-hosted release 261110 rejects language_hint on models that don't support it, instead of ignoring it. Remove those hints first.
Oct 31, 2026
Deepgram no longer supports the dg-token header for verifying callbacks: switch to Basic Auth (October 27, 2026)
Deepgram dropped dg-token for verifying callback requests. Use Basic Auth in the callback URL, or an Extra Metadata secret for STT.
Oct 24, 2026
Deepgram Nova-3 Multilingual now covers 55 languages and adds language hints (October 23, 2026)
nova-3 with language=multi now covers 55 languages. New language_hint parameter; pin version to keep the old 10-language model.
Oct 21, 2026
Deepgram confirms every Nova-3 mid-stream keyterm update, and SDKs add sendConfigure (October 19, 2026)
Each Nova-3 Configure now gets one ConfigureSuccess or ConfigureFailure. Python 7.13.0 and JS 5.15.0 add send_configure.