Deepgram now calls Aura-2 "our previous-generation model" and Flux TTS "recommended for all new builds." A buyer who pulls Aura-2's $0.030 per 1,000 characters into a comparison with Cartesia is pricing a model Deepgram has moved past, because Flux TTS costs $0.0450, a 50% step up.
That naming gap shapes the whole pair. Cartesia's flagship is Sonic 3.6, sold through monthly credit plans. Deepgram sells three generations side by side and bills them by the character, and its published first-audio claim of "as low as 80ms" is the lowest of the three companies' figures. Bland Speech joins the pair as the option for teams that care most about how the voice sounds on a call.
Which is better, Cartesia or Deepgram Aura?
As of November 2026, Cartesia is the better fit for a team that wants 44 languages from one model and instant cloning from 10 seconds of audio on a $5 plan, and Deepgram is the better fit for a team that wants the lower published first-audio figure and more concurrent streams on a pay as you go account. Neither wins on price outright: it depends on how many characters a month a team speaks.
What these rows mean
- Rate: Deepgram bills Aura and Flux per 1,000 characters of input text. Cartesia bills credits, one per character, from a monthly allowance, with overages billed at renewal only when a team turns them on.
- First audio as published: The time from sending text to hearing the first sound, as each company states it. Deepgram's figure is "as low as 80ms at full 24kHz fidelity, even under production load", and Cartesia's is "under 90ms".
- Phone audio: Mu-law and A-law are compressed 8-bit telephony formats.
- Realism result: The Elo score from Design Arena's Audio Realism Bench results of 4 August 2026, where listeners pick which of two clips sounds more like a real person. Sonic 3.6 was released after those results and Deepgram's models weren't in them.
What does Cartesia do well, and where does it fall short next to Deepgram?
Cartesia does well on languages, cloning, and developer controls, and it falls short on how credits meter a busy line. Cartesia says Sonic 3.6 "speaks 44 languages" from one model, and an instant clone needs 10 seconds of audio on the $5 Pro plan, with a professional clone from 30 minutes of audio on the $49 Startup plan, trained free and in up to 3 hours. Pronunciation dictionaries take IPA or "sounds-like" entries by id, and word-level timestamps come back over the WebSocket and SSE endpoints.
The catch is metering. Each break tag costs a credit, and with overages off, requests stop at zero credits. Cartesia's emotion controls are in beta and work for English only.
Cartesia pricing and deployment
Plans are Free at $0 with 20,000 credits, Pro at $5 with 100,000, Startup at $49 with 1.25 million, Scale at $299 with 8 million, and Enterprise custom. Overages are opt-in at $65, $45, or $38 per million credits, and credits roll over up to twice the monthly allotment. At 10 million characters a month, Scale's 8 million credits cost $299 and the other 2 million cost $76 in overage, $375 in all.
Cartesia is SOC 2 Type II certified, with BAAs and DPAs on Enterprise, and says Sonic 3.6 can run on-premise in an air-gapped environment, in a customer's VPC, or under an OEM licence. Dedicated regional deployments in the United States, European Union, United Kingdom, India, and Australia are available to Enterprise customers.
Pros
- 44 languages from one model
- Instant cloning from 10 seconds, professional cloning trained free
- Word timestamps and pronunciation dictionaries
- On-premise, VPC, and regional deployments under enterprise contracts
Cons
- Requests stop at zero credits with overages off
- Speed and volume controls don't apply to Pro Voice Clones
- Break tags and added accents cost credits
- Emotion controls in beta, English only
What does Deepgram Aura do well, and where does it fall short next to Cartesia?
Deepgram does well on concurrency headroom, self-hosting, and a low published latency claim, and it falls short on a confusing model line and a rising price. Flux TTS runs on a WebSocket at /v2/speak, with "conversational state" that persists across turns "so tone carries forward." Deepgram says responses start "in as low as 80ms at full 24kHz fidelity, even under production load." Flux TTS takes a delivery-register dial from -2 (calm) to 2 (animated) in beta, speed from 0.5 to 1.5, and IPA pronunciation in Early Access, and Deepgram says SSML tags are stripped with a warning.
The catches sit in the product line. Deepgram sells Aura-1, Aura-2, and Flux TTS together, with Aura-2 described as previous-generation and Aura served on a different endpoint, /v1/speak, than Flux on /v2/speak. Flux costs $0.0450 per 1,000 characters on pay as you go and Growth alike, where Aura-2 costs $0.030 and $0.027. Deepgram says Flux ships in English plus 10 languages with built-in text normalization and supports 40+ more without it, and Deepgram offers 42 language codes for Flux TTS, 32 of them in Early Access.
Deepgram pricing and compliance
Deepgram's plans are Pay as you go with no minimum, Growth from $4,000 a year prepaid, and Enterprise custom. A free account gets a $200 credit with no card and no expiry, and through December 31, 2026 Flux TTS spend earns $1 in credits for every $1 spent, up to $500. Growth overage is billed at the Growth rate plus 10%, weekly. At 10 million characters a month, Flux TTS on pay as you go costs $450.
Deepgram says it is SOC 2 Type 1 and Type 2 certified, HIPAA compliant, and PCI compliant, signs Business Associate Agreements for Enterprise customers handling electronic protected health information, and offers a dedicated EU endpoint. Flux TTS supports self-hosted deployment in a customer's own cloud, a private cloud, or on-premise.
Pros
- Streaming up to 45 concurrent on pay as you go in North America
- Published first audio of as low as 80ms
- Self-hosted, private cloud, and on-premise options
- $200 free credit with no card
Cons
- Three model generations on two endpoints
- Flux TTS at $0.0450 per 1,000 characters, up from Aura-2's $0.030
- 32 languages in Early Access
Deepgram's Aura line is a poor fit for a team that wants a single current model name to build against, since Flux TTS is the one Deepgram recommends.
Which sounds more human, Cartesia or Deepgram Aura?
On Design Arena's Audio Realism Bench results of 4 August 2026, Cartesia's Sonic 3.5 ranked 6th of 13 AI voices at 1096, and Deepgram wasn't in those results, so only Cartesia has a score to read. Elo is a rating that rises each time listeners pick a voice as more human.
- #1BlandBland Speech v31365
- 2ndMicrosoftMAI-Voice-21224
- 3rdxAIGrok TTS1156
- 4thOpenAIGPT Realtime 21119
- 5thGoogleGemini 2.5 Pro TTS1109
- 6thCartesiaSonic 3.51096
Ranks are among 13 AI voices, and Cartesia's own voice is on the list. Real human recordings scored 1500, at the right edge of this scale. Bars start at 600, not zero. Source: Design Arena's published results.
Cartesia released Sonic 3.6 on 27 August 2026, after the results, so even Cartesia's score belongs to its previous model. Bland Speech v3 ranked 1st at 1365 and the human recording scored 1500, in a test where, as Design Arena's Audio Realism Bench article describes, vetted native speakers pick the more human of two recordings of one script.
Where does Bland Speech sit next to Cartesia and Deepgram Aura?
Bland Speech sits between the two on price, and Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, at 1365. Bland bills one standard rate with no monthly plan. Bland Speech costs $0.04 per 1,000 characters, pay as you go, with 50,000 free characters once for a new account, which puts $40 per million between Aura-2's $30 and Flux TTS's $45. At 10 million characters a month that is $400, a little above Cartesia's $375 on Scale.
Bland publishes 199ms to first audio, higher than Cartesia's under 90ms and Deepgram's 80ms, and a team that ranks on that number alone will pick one of them. Bland Speech streams over HTTP chunked and a WebSocket where a new context_id interrupts a playing turn, outputs mu-law at 8 kHz or PCM from 8 to 48 kHz, and accepts performance tags such as [laughs] and [say warmly]. An instant clone needs one clean sample of about 10 seconds, and a professional clone is fine-tuned on 30 minutes or more of audio and goes through Bland's consent and rights review. Pipecat and LiveKit Agents both have a Bland Speech integration, and the text to speech API page lists the endpoints.
Bland is SOC 2 Type II audited and HIPAA compliant with a signed BAA. Enterprise customers get dedicated orchestration and dedicated GPUs run by Bland, and Enterprise pricing is custom, set with Bland's sales team.
Pros
- Ranked 1st of 13 in the 4 August 2026 realism results
- One current model name and one standard rate
- Barge-in built into the WebSocket protocol
- Pipecat and LiveKit integrations
Cons
- More than a prototype needs, because Bland Speech is built for production calls
The Bland Speech vs Deepgram Aura page sets Deepgram against Bland, best Cartesia alternatives and best Deepgram Aura alternatives cover the exits, and the text to speech hub and rival vs rival family hold the other pairs. Bland Speech has the studio.
Frequently asked questions
Is Deepgram Aura still Deepgram's current text to speech model?
No. Deepgram calls Flux TTS its "latest-generation, conversation-native voice model" and "recommended for all new builds", and describes Aura-2 as its previous-generation model. Aura-2 and Aura-1 are still sold and served on /v1/speak.
Is Cartesia or Deepgram faster?
Deepgram publishes the lower number: first audio "as low as 80ms" for Flux TTS against Cartesia's "under 90ms" for Sonic. Each company states its figure its own way, and neither is a race Bland ran. Bland publishes 199ms to first audio.
Which is cheaper, Cartesia or Deepgram?
Cost depends on volume. Cartesia's Startup plan costs $49 for 1.25 million characters, and Deepgram's Flux TTS costs $45 per million characters on pay as you go. At 10 million characters a month, Cartesia's Scale plan with overage comes to $375 and Flux TTS to $450. Bland Speech is $40 per million characters.
Which handles more simultaneous calls, Cartesia or Deepgram?
Deepgram handles more on a self-serve account in North America. Deepgram allows Flux TTS streaming of up to 45 on pay as you go and 60 on Growth, against Cartesia's TTS limits of 3 on Pro, 5 on Startup, and 15 on Scale. Outside North America, Deepgram's Flux TTS streaming limit is 5.
Do Cartesia and Deepgram support phone audio formats?
Yes. Cartesia offers mu-law and A-law with sample rates from 8,000 to 48,000 Hz, and Deepgram offers mu-law and A-law at 8,000 Hz by default or 16,000 Hz. Bland Speech offers mu-law at 8 kHz and PCM from 8 to 48 kHz.
Does Cartesia support more languages than Deepgram?
Cartesia says Sonic 3.6 speaks 44 languages. Deepgram says Flux TTS ships in English plus 10 additional languages with built-in text normalization and supports 40+ more without it, and Deepgram's Aura-2 voices span seven languages.
