Both companies build for real-time voice, but Deepgram sells a pay-as-you-go stack that covers batch and streaming recognition, synthesis, agents, and self-hosting, while Cartesia sells monthly credit plans centered on streaming speech-to-text, text-to-speech, and managed voice agents.

Introduction and methodology note

Deepgram and Cartesia both target developers building conversational voice products. Each sells streaming speech-to-text with built-in turn detection, text-to-speech tuned for low latency, and a hosted path to a working voice agent. Both offer free entry points and publish self-serve prices.

They differ in packaging and scope. Cartesia bills through monthly plans (Free, Pro, Startup, Scale) that grant a pool of credits shared across Sonic text-to-speech and Ink speech-to-text, plus a separate per-minute rate for managed agents. Deepgram bills pay-as-you-go per minute, character, or token, with an optional prepaid Growth plan, and adds pre-recorded transcription, audio intelligence, and self-hosted deployment.

Methodology note: Deepgram figures come from Deepgram's own published pages, linked inline. Cartesia figures come from Cartesia's own published pricing and product pages and are stated as plain text. Both were checked on September 30, 2026. Cartesia effective rates in this page are derived from its published plan prices and included quantities, and the derivation is shown. Where a vendor does not publish a number, this page says "not published." Latency and accuracy claims are self-reported, and testing on your own audio is the only reliable comparison.

TL;DR

• Choose Deepgram for pre-recorded and streaming transcription in one API, usage-based billing with no monthly plan, composable agent pricing from $0.075/min down to $0.050/min, and self-hosted deployment.
• Choose Cartesia for a latency-focused streaming stack with bundled instant and professional voice cloning, a low-cost entry plan ($5/month), and managed agents at a flat $0.06/min call rate.
• At Cartesia's Scale plan, effective streaming speech-to-text is about $0.40/hr, which falls between Deepgram's promotional ($0.288/hr) and regular ($0.462/hr) Nova-3 streaming rates.

Quick comparison table

Granular differentiator table: plan economics

Cartesia's credit plans are its defining commercial difference. The table converts each plan into an effective rate at 1 credit per character for Sonic-3.6 and 3 credits per second for Ink-2, assuming all credits are used.

Deepgram's pay-as-you-go comparison points are Aura-2 at $30 per 1M characters, Flux TTS at $45 per 1M characters, and Nova-3 streaming at $0.288/hr promotional or $0.462/hr regular. Aura-2 is about 20% below Cartesia's Scale effective TTS rate. Flux TTS is above Startup and Scale effective rates and below the Pro rate. Cartesia's plans favor teams with steady monthly volume that fills the included credits, and Deepgram's usage billing favors spiky or early-stage volume because nothing is prepaid. Overage rates are not shown on the Cartesia pricing page, so model overage before committing to a plan.

Why teams choose Deepgram over Cartesia

Batch and streaming recognition in one API

Deepgram covers pre-recorded transcription, real-time meeting and caption streams, and turn-based agent streams in one set of endpoints (streaming feature overview). Nova-3 pre-recorded runs $0.0043/min with diarization included. Cartesia presents Ink as a streaming API and does not publish a pre-recorded rate, so archive and call-recording workloads need a second vendor.

Composable agent pricing and agent-oriented synthesis

Deepgram publishes agent tiers from $0.075/min Standard to $0.050/min when you bring your own LLM and TTS. Flux TTS is built for the agent loop, with a real-time WebSocket, a REST transport for fixed audio, and interruption reporting that states what the user heard (Flux TTS overview). Cartesia's $0.06/min is a lower headline, but LLM usage during calls is described as free for agents created in its UI for a limited time, so compare total cost once that period ends.

Usage billing and published concurrency

Deepgram has no monthly subscription. A project pays for what it uses, starting from a $200 credit. Published concurrency limits (up to 150 WebSocket connections on pay-as-you-go and 225 on Growth for speech-to-text) are visible before signup (getting started guide). Cartesia's Pro plan includes 12 concurrent Ink-2 requests and 3 concurrent Sonic requests, which bounds small production workloads.

If a single API for batch, streaming, and agents fits your roadmap, start a free trial with $200 in Deepgram credits.

Best use cases for Deepgram

Real-time voice agents with flexible components

Flux provides model-integrated end-of-turn detection, and the Voice Agent API lets teams swap in their own LLM and TTS at lower per-minute tiers.

How Cartesia compares: Ink-2 also signals turn start and turn end natively and adds an early end-of-turn signal, and its managed agents carry a flat $0.06/min call rate.

Call recordings and meeting archives

Pre-recorded Nova-3 with included diarization and smart formatting suits contact-center and meeting transcription at $0.0043/min.

How Cartesia compares: It does not publish a pre-recorded speech-to-text product.

Regulated and self-hosted deployments

Self-hosted speech-to-text, text-to-speech, and Voice Agent deployment, HIPAA BAAs on Enterprise, and SOC 2 Type 1 and 2 are published.

How Cartesia compares: It lists HIPAA, SOC 2 Type 2, GDPR, and PCI, with BAAs on Enterprise, and references local deployment without publishing terms.

Spiky or early-stage volume

Pay-as-you-go billing with a $200 starting credit avoids paying for unused monthly credits.

How Cartesia compares: Its $5 Pro plan is a cheap way to start, and its credit pool is cost-effective once monthly volume reliably fills it.

Best use cases for Cartesia

Latency-first voice agents with bundled cloning

Cartesia designs Sonic and Ink for real-time conversation and reports a time to final transcript of 0.1 seconds for Ink. Instant voice cloning starts at $5/month and professional cloning starts at $49/month with free training. Teams that want a distinctive branded voice and a single vendor for voice models and managed agents fit this profile.

Predictable monthly voice budgets

A team with steady volume can buy a plan, read the included minutes off the pricing page, and forecast cost as a flat monthly number. At Scale, $299 covers about 10,667 text-to-speech minutes or about 740 hours of Ink-2.

Platform overview of Deepgram

Speech-to-text with Nova-3

Nova-3 serves pre-recorded and streaming audio in 45+ languages with diarization, keyterm prompting, and language detection. See the models and languages overview.

Flux conversational recognition

Flux is a streaming model with built-in end-of-turn detection for agents, in English and multilingual variants. Start with the Flux quickstart.

Text-to-speech with Aura-2 and Flux TTS

Synthesis is sold standalone at $0.015 to $0.045 per 1k characters. The text-to-speech guide covers requests and voices.

Voice Agent API

A single WebSocket handles listening, reasoning, and speaking, with bundled and bring-your-own tiers. Read the Voice Agent documentation.

Audio Intelligence

Summarization, topic detection, sentiment, and intent recognition are token-priced, with entity detection and redaction billed per minute. See the Audio Intelligence product page.

Integrations

Deepgram documents LiveKit, Pipecat, Twilio, Amazon Connect, and Genesys integrations. The LiveKit integration guide shows a working agent pattern.

How to choose

Switching from Cartesia to Deepgram

1. Set up the account. Create a Deepgram project and key, and use the $200 credit for a side-by-side test. The Deepgram Python SDK is the fastest way to a first streaming request.
2. Swap streaming recognition. Ink WebSocket sessions map to Flux for agents or Nova-3 for general streaming. Cartesia's turn.start and turn.end events map to Flux end-of-turn events, so review your turn-handling code rather than assuming a one-to-one rename.
3. Swap synthesis. Sonic voices map to Aura-2 or Flux TTS voices. Cloned voices do not transfer, and Deepgram does not publish voice cloning on its pricing page, so confirm availability before moving a cloned voice.
4. Rework cost tracking. Credits and plan allotments become per-minute and per-character line items. Model a month of real traffic against both billing shapes.
5. Run in parallel. Mirror one to two weeks of traffic, and compare word error rate, time to final transcript, and interruption behavior on your own calls.
6. Move agents last. If you use Cartesia Managed Agents, rebuild the loop on the Voice Agent API or on LiveKit or Pipecat with Deepgram components, and compare all-in cost including the LLM.

FAQ

Is Deepgram cheaper than Cartesia?

For streaming speech-to-text, Deepgram is lower at the promotional Nova-3 rate ($0.288/hr against an effective $0.40/hr on Cartesia Scale) and higher at the regular rate ($0.462/hr). Aura-2 is about 20% below Cartesia's Scale effective text-to-speech rate. Cartesia's agent call rate of $0.06/min is lower than Deepgram's $0.075/min Standard tier.

Which has lower latency?

Both publish self-reported figures and neither figure is directly comparable. Cartesia reports 0.1 seconds to final transcript for Ink. Measure time to final transcript on your own telephony audio with both.

Does Cartesia offer batch transcription?

Cartesia presents Ink as a streaming speech-to-text API and does not publish a pre-recorded rate on the pages reviewed.

Can I self-host either platform?

Deepgram publishes self-hosted deployment for Enterprise customers. Cartesia lists on-premises and VPC deployment as a topic in its pricing FAQ without terms on the pages reviewed, so confirm with its sales team.

Does Deepgram offer voice cloning?

Voice cloning is not published on Deepgram's pricing page. Cartesia includes instant cloning from its Pro plan.

How hard is migration?

Recognition moves in days because both use WebSocket streaming with SDKs. Plan extra time for voice selection and for replacing Cartesia plan billing with usage billing.

Both vendors offer a free entry point, so the fastest way to settle this comparison is a test on your own calls. When you are ready to size it for production volume, talk to the Deepgram sales team about Growth and Enterprise plans.

Related Articles

• Deepgram vs. Azure
• Deepgram vs. OpenAI