What does the Voice Agent API cost include for LLM and other Deepgram APIs?

The Voice Agent API pricing is based on WebSocket connection time , which includes usage of Deepgram's speech-to-text (STT) and text-to-speech (TTS) models—there’s no separate charge for these during a session. If you choose to use Deepgram's built-in LLM support (currently via OpenAI or Anthropic ), that’s also included in the price. You can optionally use your own LLM instead, which may reduce WebSocket time (and thus cost), depending on your setup. At the moment, Deepgram doesn’t offer pricing tiers based on model selection for TTS/STT, but you can configure specific model options in your Agent Settings Configuration if you prefer something other than the defaults.

How is multichannel billed?

When you opt into using the multichannel feature, each channel is transcribed and billed separately. The total cost when using multichannel is the single-channel cost multiplied by the number of channels. When you do not enable multichannel, Deepgram converts your multichannel audio into mono single-channel audio, and it is transcribed and billed as one channel. We especially recommend using the multichannel feature on multichannel audio in which there is cross-talk (voices overlapping or talking over each other) for the most accurate transcription and speaker detection.

What's the difference between Nova, Enhanced and Base models?

Nova is our newest and most powerful model, offering the best balance between accuracy and cost-effectiveness. Enhanced is a powerful ASR model that performs especially well with uncommon words. Base is our signature model, with a solid combination of accuracy and cost-effectiveness. Some languages are only supported by Enhanced and Base. See our Models Overview and Model docs.

Which file types can you transcribe?

We support over 40 audio and video formats, documented here .

How does billing work?

You purchase credit upfront with a credit card. Credit will be deducted from your balance as you use our API. Pay As You Go credit never expires. Growth plan credit expire 1 year from purchase unless you renew or upgrade.

What unit of time is billed, minutes or seconds?

Deepgram bills by the second of audio. For instance, if you transcribe 61 seconds of audio, we bill you for 61 seconds of usage, not 2 minutes (120 seconds).

Can you transcribe live streaming audio?

Definitely. In fact, we’ve got the fastest real-time transcription in the biz with latency times of under 300 milliseconds.

Can Deepgram transcribe real-time conversations?

Yes! Our streaming API is designed for low latency and will return incremental transcripts as a speaker’s sentence unfolds. You can stay up to date on our latest Conversational AI technology by subscribing to our newsletter .

What happens if I run out of credit before my Growth plan expires?

If you’re on the Growth plan and have saved a credit card, you can continue to use our API with a 10% overage fee billed at the start of each month. This is still less than Pay As You Go rates. You can renew or upgrade your plan at any point to prevent an overage fee.

What languages do you support?

We support 36+ languages with Speech to Tex t and English for Text to Speech. We are actively increasing the number of languages available for our products.

Can I get human support?

Sure thing. You can get help from our community over at Discord or GitHub Discussions . If you are subscribed to our Enterprise plan, reach out to our support team via email or Slack for assistance.

How many seats do i get?

We bill based on usage, not users. Add as many team members and collaborators as you wish!

Can i self-host Deepgram?

Yes, our Enterprise plan offers self-hosted deployments of our voice AI products, for use in cloud environments or on-premises datacenters. Contact us about getting on an Enterprise plan to expand your deployment capabilities.

Are the Audio Intelligence features affected by the model I select?

Our Audio Intelligence features such as topic detection and summarization are not dependent on the selected model. However, audio intelligence features are primarily affected by the transcript, so you should use the most accurate model available for your use case for best results.

Is streaming available for Deepgram TTS Aura?

Streaming is available on our Live TTS service over WebSocket.

How can I increase my concurrency and requests per minute for Aura TTS?

Higher concurrency and requests per minute is available. Please contact sales .

Is voice cloning available for TTS Aura?

Voice cloning is currently not available for Aura but we are looking into voice cloning in the future.

The Deepgram Enterprise Voice AI accelerator program

Build enterprise-grade voicebots, AI agents, and apps fast and easy at scale. Sign up now!

Sign Up Learn More

Unlock exclusive benefits in our expanded Voice AI accelerator

Access to Deepgram's experts for system design and model customization tailored to your use case
Faster time to market with the hands-on support you need to quickly move from POC all the way to full production at scale
Exclusive early access to unreleased voice AI innovations
Discounts and priority placement in developer workshops and community events like our upcoming Voice AI Agents Workshop
Free developer credits and access to Deepgram’s pre-production environments for development and testing support

Frequently Asked Questions

Are you ready to innovate?

Partner with us to unlock new opportunities and elevate your business with voice AI.