models/Deepgram/Nova-3
Deepgram

Nova-3

audio-stt
Compare

Transcribe audio using Deepgram’s speech-to-text model

MODALITIES
audio
RELEASED
2025-06-05

Nova-3 (deepgram/nova-3) is a audio-stt model from Deepgram, released 2025-06-05. Pricing via AIgateway: $0.0052 per minute. Capabilities: streaming, async. Call it via https://api.aigateway.sh/v1/audio/transcriptions — set model="deepgram/nova-3". Best for: Call-center transcription, Voice agents, Meeting transcription.

model · deepgram/nova-3family · Nova

Use this model

model: deepgram/nova-3
curl https://api.aigateway.sh/v1/audio/transcriptions \
  -H "Authorization: Bearer $AIGATEWAY_API_KEY" \
  -F model="deepgram/nova-3" \
  -F file="@audio.mp3"

Async & streaming

Async transcription — submit and poll /v1/jobs/<id>, or have the result pushed to your webhook_url. Best for long files and batch pipelines.
# Submit (returns immediately with a job id)
curl -X POST https://api.aigateway.sh/v1/audio/transcriptions \
  -H "Authorization: Bearer $AIGATEWAY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"deepgram/nova-3","audio_url":"https://example.com/audio.wav","async":true}'
# -> {"id":"<job_id>","status":"processing"}

# Poll for the transcript
curl https://api.aigateway.sh/v1/jobs/<job_id> \
  -H "Authorization: Bearer $AIGATEWAY_API_KEY"

# ...or skip polling: pass "webhook_url" and we POST the signed result when ready
#   {"model":"deepgram/nova-3","audio_url":"...","webhook_url":"https://you.example.com/hook"}
Realtime streaming — open a WebSocket, stream audio frames, and receive interim + final transcripts live. Billed per minute at the websocket rate (higher than batch).
// Realtime WebSocket. Browsers pass the key as ?api_key=
const ws = new WebSocket(
  "wss://api.aigateway.sh/v1/realtime?model=deepgram/nova-3&encoding=linear16&sample_rate=16000&interim_results=true&api_key=" + AIGATEWAY_API_KEY,
);
ws.onmessage = (e) => {
  const msg = JSON.parse(e.data);
  if (msg.type === "Results") {
    console.log(msg.channel.alternatives[0].transcript, msg.is_final);
  }
};
// stream raw audio frames (linear16 PCM @ 16 kHz):
// ws.send(pcmChunk)
// ...then end the stream:
ws.send(JSON.stringify({ type: "CloseStream" }));

Capabilities

StreamingAsync

Strengths

  • Real-time streaming
  • 95%+ accuracy on noisy audio
  • 40+ languages

Use cases

Call-center transcriptionVoice agentsMeeting transcription

Pricing

Per minute$0.0052
Batch $0.0052/min · WebSocket streaming $0.0092/min
You pay pass-through · 5% applied at credit top-up, not per-call.
See API example →CompareAPI referenceSee usage ranking →

Collections

More audio models →More from DeepgramFrontier models →Free-tier models →
API schema

Call Nova-3 from any OpenAI SDK

POST https://api.aigateway.sh/v1/audio/transcriptions·Content-Type: multipart/form-data·Auth: Bearer sk-aig-...

Request body

json
# multipart/form-data — use curl -F or SDK file upload
model="deepgram/nova-3"
file=@audio.mp3
response_format=json    # or "verbose_json", "text", "srt", "vtt"
language=en             # optional

Response

json
{
  "text": "Hello from AIgateway.",
  "language": "en",
  "duration": 1.82
}

Quickstart

from openai import OpenAI
client = OpenAI(base_url="https://api.aigateway.sh/v1", api_key="sk-aig-...")

with open("audio.mp3", "rb") as f:
    r = client.audio.transcriptions.create(model="deepgram/nova-3", file=f)
print(r.text)

Errors

401authentication_errorInvalid or missing API key
402insufficient_creditsWallet empty (PAYG only)
404not_foundUnknown model or endpoint
429rate_limit_errorOver per-minute limit — see Retry-After header
500server_errorUpstream provider failed (retryable)
503service_unavailableUpstream saturated (retryable)
Full docs →API reference →OpenAPI spec →llms.txt →

Frequently asked questions

What is Nova-3?
Transcribe audio using Deepgram’s speech-to-text model It is a audio-stt model from Deepgram, accessible via AIgateway's OpenAI-compatible API at slug deepgram/nova-3.
How much does Nova-3 cost via AIgateway?
$0.0052 per minute of audio. Pass-through plus a 5% platform fee applied at top-up.
How do I call Nova-3 from my code?
Point the OpenAI SDK at https://api.aigateway.sh/v1 with your AIgateway key and set model to "deepgram/nova-3". The request and response shapes match OpenAI exactly.
Does Nova-3 support streaming, tool calling, vision, and JSON mode?
Streaming — yes. Tool calling — no. Vision — no. JSON mode — no. Prompt caching — no.
What are the best use cases for Nova-3?
Call-center transcription, Voice agents, Meeting transcription. Key strengths: Real-time streaming; 95%+ accuracy on noisy audio; 40+ languages.
Can I bring my own Deepgram API key (BYOK)?
Yes. Attach a Deepgram key in your AIgateway dashboard and this model flips to pass-through — you pay Deepgram directly and AIgateway waives the 5% platform fee on those calls.