Deepgram pricing in 2026: Nova-3 cost per minute vs Whisper

How much does Deepgram speech to text cost?

Deepgram bills speech to text per second of audio, with no rounding up to a block and no minimum. Multichannel audio bills by the total duration across channels, so a 10-minute stereo file bills as 20 minutes. Figures come from Deepgram’s pricing page. Prices checked November 10, 2026.

Growth is a prepaid annual plan from $4,000 a year that cuts rates by up to 20%. Older models keep their rates for existing deployments: Nova-2 streaming at $0.35 an hour, Enhanced at $0.99 and Base at $0.87.

Which Deepgram features cost extra?

Smart formatting is free everywhere, and speaker diarization is free on pre-recorded audio. Other features add a per-minute charge on top of transcription. Summaries, topic detection, sentiment and intent recognition bill $0.0003 per 1,000 input tokens and $0.0006 per 1,000 output tokens.

What do Deepgram text to speech and Voice Agent cost?

Text to speech bills per 1,000 characters: Aura-2 at $0.030, Aura-1 at $0.015 and Flux TTS at $0.045. Through December 31, 2026, Deepgram adds $1 of credit for each $1 spent on Flux TTS, up to $500.

Voice Agent bills per minute of connection time and covers speech to text, a managed language model and text to speech. Standard costs $0.075 a minute ($4.50 an hour). Bringing your own language model and voice drops it to $0.050 a minute. Advanced costs $0.163 a minute.

What does the free credit cover, and how do plans work?

Pay as you go starts with $200 of credit that does not expire and needs no card. Deepgram says that covers about 43,000 minutes, over 700 hours, of Nova-3 transcription. After that you pay per use with no minimum.

Growth customers who run past their prepaid credit pay their Growth rate plus 10%, billed weekly. Pay as you go allows 150 concurrent streaming connections and Growth allows 225. Deepgram reviews refund requests for unused credits bought in the last 30 days.

How does Whisper API pricing compare with Deepgram?

Whisper is OpenAI’s speech model. You can buy it from OpenAI, from Deepgram, from Groq, or run the open-source model yourself. OpenAI’s whisper-1 API is deprecated and shuts down on February 26, 2027. OpenAI now points new work to gpt-transcribe. The table prices 100 hours of pre-recorded audio.

• Deepgram vs Whisper: features and accuracy trade-offs

What does Locus Pro add for Deepgram?

Locus Pro offers Deepgram transcription as deepgram/transcribe, paid from one prepaid credit balance that also covers other providers, with no Deepgram account. It takes a public audio URL and defaults to Nova-3, with language=multi and Whisper Large as options. Each call bills the audio Deepgram reports, per second. Redaction, entity detection, summaries, topics, sentiment and intents are available and add to the charge.

Locus holds up to 4 hours of audio when the call starts and refunds the rest once the transcript returns. Your agent sees the price first and can cap any call, and a monthly hard limit stops new calls. For a local file, the endpoint suggests buying an upload slot from stableupload/upload, an x402 listing Locus has not reviewed.

• Transcribe audio cheaply
• Browse audio endpoints

Which transcription option should you choose?

For steady volume, use your own Deepgram account and API key: billing is per second, diarization is free, you start with $200, and transcription at that volume can cost less per minute. If price per hour is all that matters and you can live without speaker labels, Groq’s Whisper Turbo is far cheaper.

Locus Pro fits agents that transcribe now and then, next to search, scraping and models on one bill, without a Deepgram account. Each call can carry a price cap, and on Enterprise you can bill your own customers for the audio their agents transcribe.

• Live captions or voice agents: Deepgram streaming at $0.0077 a minute.
• Meetings and calls with several speakers: Deepgram Nova-3 pre-recorded, diarization included.
• Bulk archives on a budget: Groq whisper-large-v3-turbo at $0.04 an hour.
• Existing whisper-1 code: plan a move before February 26, 2027.