How much does Deepgram cost per minute?

Speech-to-text rates for pre-recorded and streaming audio

Your base transcription bill comes down to two choices: whether you stream audio over a WebSocket or upload finished audio files, and whether you run monolingual or multilingual models. Pre-recorded files processed through the batch API are cheaper across the board.

For pre-recorded audio on Pay as you go, Nova-3 Monolingual costs $0.0043 per minute, while Nova-3 Multilingual costs $0.0052 per minute. On the Growth plan, those drop to $0.0036 and $0.0043 per minute respectively. If you run Whisper Large through Deepgram's pre-recorded API, the rate stays locked at $0.0048 per minute whether you're on Pay as you go or Growth.

Streaming transcription carries a slight premium because servers must hold open connections and return words with low latency. Streaming Nova-3 Monolingual and Flux STT English cost $0.0077 per minute on Pay as you go, or $0.0065 per minute on Growth. Multilingual streams cost more: Flux STT Multilingual runs $0.0078 per minute on Pay as you go ($0.0068 on Growth), and Nova-3 Multilingual costs $0.0092 per minute on Pay as you go ($0.0078 on Growth).

If your stack still relies on older model generations, you aren't forced to migrate immediately. Deepgram keeps legacy models available at their original Pay as you go streaming prices:

• Nova-2 streaming costs $0.35 per hour ($0.00583 per minute).
• Enhanced streaming costs $0.99 per hour ($0.0165 per minute).
• Base streaming costs $0.87 per hour ($0.0145 per minute).
• Growth plan accounts pay about 12.5% less for these legacy models.

Multichannel audio multiplies your cost. If you send a two-channel stereo recording that lasts ten minutes, Deepgram bills you for twenty minutes of processed audio.

Add-ons, Text-to-Speech, and Voice Agent costs

Turning on optional processing toggles changes your per-minute bill. Some toggles carry an extra per-minute surcharge, while others are built directly into the base engine.

Transcription add-ons

Smart Formatting is included at no extra cost across all plans. Speaker Diarization is also included for free on pre-recorded audio, but if you enable it on a live stream, it adds $0.0020 per minute on Pay as you go or $0.0017 on Growth. Redaction costs $0.0020 per minute on Pay as you go and $0.0017 on Growth for both streaming and pre-recorded files. Keyterm Prompting adds $0.0013 per minute on Pay as you go and $0.0012 on Growth. Entity Detection runs $0.0017 per minute on both plans.

If you request Audio Intelligence features on top of transcription - such as Summarization, Topic Detection, Sentiment Analysis, or Intent Recognition - Deepgram bills them by token usage rather than audio minutes. On Pay as you go, input tokens cost $0.0003 per 1k and output tokens cost $0.0006 per 1k. On Growth, those rates drop to $0.00024 per 1k input and $0.00048 per 1k output tokens.

Text-to-Speech (TTS)

Text-to-Speech does not bill by audio duration at all; it bills based on the character length of the text you supply.

• Aura-1 costs $0.0150 per 1,000 characters on Pay as you go and $0.0135 on Growth.
• Aura-2 runs $0.030 per 1,000 characters on Pay as you go and $0.027 on Growth.
• Flux TTS costs $0.0450 per 1,000 characters on both plans. Deepgram has a promotion running through December 31, 2026, where they credit $1 for every $1 spent on Flux TTS up to $500.

Voice Agent API

The Voice Agent API bills for raw WebSocket connection time. Standard Voice Agent costs $0.075 per minute ($4.50 per hour) on Pay as you go, and $0.068 per minute ($4.08 per hour) on Growth. If you bring your own components, those base connection fees decrease:

• Standard with bring-your-own (BYO) TTS costs $0.065 per minute on Pay as you go ($0.051 on Growth).
• Custom with BYO LLM costs $0.065 per minute on Pay as you go ($0.059 on Growth).
• Custom with BYO LLM and TTS costs $0.050 per minute on Pay as you go ($0.041 on Growth).
• Voice Agent Advanced costs $0.163 per minute ($9.78 per hour) on Pay as you go and $0.146 on Growth. With BYO TTS, Advanced runs $0.122 per minute on Pay as you go and $0.110 on Growth.

Pay as you go vs Growth plans

Small projects generally sit on Pay as you go, where you maintain a cash balance or pay against a card as you burn audio. Larger teams commit to the Growth tier, which starts at $4,000+ per year in pre-paid credits and yields up to 20% in per-minute savings.

Growth plans introduce strict overage mechanics. If your team exhausts its pre-paid credits, your traffic doesn't get shut off, but Deepgram bills overages at your Growth rate plus a 10% penalty. Overages are tallied and charged weekly on Mondays for the prior seven days. You must keep an active card on file; if credits run dry and no card exists, every API call returns an HTTP 402 error immediately. You can avoid that 10% penalty by early renewing a new Growth plan inside the console before your balance drops below zero, which resets your credits back to the standard Growth rate and shifts your annual billing date forward.

If you purchase credits on either plan and change your architecture, unused credits purchased within the last 30 days can be refunded by emailing consolebilling@deepgram.com.

Project balances, API key quirks, and concurrency limits

When you sign up, Deepgram creates your primary project automatically and assigns any promotional balance directly to it. If you build separate projects to isolate staging from production, credits don't pool together. You have to email Support to transfer credit balances between projects, and whatever expiration date was attached to those credits will transfer right along with them.

There is a documentation discrepancy around expiration: the official pricing page states that the initial $200 free credit does not expire, while the project management guide states that free promotional credits expire one year from signup. Credits purchased by individuals using a credit card do not expire, whereas enterprise contract credits terminate when the contract period ends.

Rate limits apply strictly per project rather than per user account or per API key. Self-serve secondary projects are restricted to a single concurrent stream by design. If you try to spread production load across multiple secondary projects under the same account to get around throughput ceilings, you run straight into that 1-stream wall and violate Deepgram's terms of service.

Regional endpoints matter for both latency and concurrency:

• North America: api.deepgram.com
• Europe: api.eu.deepgram.com
• Australia: api.au.deepgram.com
• India: api.in.deepgram.com

On Pay as you go across all four regions, speech-to-text limits sit at up to 50 concurrent pre-recorded requests and up to 150 concurrent streams for Nova-3, Nova-2, Nova, Enhanced, and Base. Flux STT also caps at 150 concurrent streams. On Growth, North American streaming concurrency bumps to 225 concurrent requests, while Europe, Australia, and India remain at 150. Whisper Cloud pre-recorded processing is only available on the North American endpoint, limited to 3 concurrent requests on both Pay as you go and Growth.

Adding certain features pulls your concurrency limits down. When you combine services in a single request - like running Speech-to-Text alongside Sentiment Analysis - Deepgram enforces whichever service has the lower rate limit. Enabling Speaker Diarization on Pay as you go or Growth caps your requests at 50 pre-recorded and 50 streaming in North America, and cuts you down to 25 pre-recorded and 25 streaming requests in Europe, Australia, and India.

For API key hygiene, programmatic key creation is limited to 250 keys per day. If you remove the team member who generated an API key from the project or delete their user account, that key stops working immediately. You can append query parameters like tag=apple to individual API requests or tag keys directly, letting you parse downstream usage inside project reports.

Tracking Deepgram balance drops with Costbrake

Deepgram projects stop processing the moment your pre-paid balance hits zero, unless you're on Growth with a credit card attached. Because Deepgram does not offer native read-only keys, monitoring your balance programmatically requires a key with Owner or Admin permissions.

Costbrake tracks your Deepgram balances alongside your other infrastructure. It works with cloud, AI and SaaS providers including AWS, Google Cloud, Azure, Cloudflare, Vercel, OpenAI, Anthropic and Deepgram. Every check reads the current USD balance across up to 25 projects visible to your key and tracks the drop in funds between cycles. Projects billed by monthly invoice don't carry a running USD balance, so this tracking applies strictly to balance-backed accounts.

Checks run hourly on Costbrake's Free plan and every 5 minutes on Pro and Team, accelerating to 2-minute intervals whenever a spend spike is detected. A spike alert fires if an hour's usage exceeds three times your 7-day hourly baseline and costs at least $5 more than usual. You also get automated budget threshold alerts at 50%, 80%, and 100% of your targets, along with month-end forecast warnings when your trailing 72-hour burn rate threatens to blow past your limit. Alerts route to email on Free, expanding to Slack, Discord, and signed webhooks on Pro, and PagerDuty on Team.

Unlike other providers where Costbrake offers opt-in, reversible brakes that an admin can click to pause workloads, Deepgram API keys cannot be paused; they can only be deleted. If an API key leaks, head to the Deepgram console immediately to delete it, or review what should I do in the first hour after an API key leaks? for triage steps.

You can set up balance tracking by having your AI assistant run the agent setup flow, or connect your project manually at https://app.costbrake.com/signin. For details on tier limits and team seats, see the Costbrake pricing page.