A buyer searching for Deepgram Aura is searching for a model Deepgram now calls previous-generation. Deepgram describes Aura-2 as "Our previous-generation model" and says its newer Flux TTS is "Recommended for all new builds", at $0.045 per 1,000 characters against $0.030 for Aura-2 and $0.015 for Aura-1. A comparison with ElevenLabs therefore has three Deepgram prices to hold in mind, and the cheapest is the oldest voice.
A team weighing the two for phone calls should also look at Bland Speech, which costs $0.04 per 1,000 characters with no plan. Bland Speech v3 ranked 1st of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026. The text to speech comparison hub places every vendor, and the rival comparisons hold the other pairs.
Which is better, ElevenLabs or Deepgram Aura?
Deepgram is the better pick for a team that already uses Deepgram for speech recognition, wants a stated 80ms latency claim, and may host the model itself, as of November 2026. ElevenLabs is the better pick for a team that wants audio tags for emotion, cloning from a $6 plan, and 90+ languages on its v4 models.
Ten million characters a month costs $450 on Deepgram's Flux TTS, $300 on Aura-2, $150 on Aura-1, $400 at ElevenLabs' v4 Turbo rate, and $800 at its v4 rate, before plan fees and credits.
What these rows mean
- Audio Realism Bench, 4 August 2026: Design Arena's blind listening test, where native speakers hear two recordings of the same line and pick the one that sounds more like a real person. ElevenLabs released Eleven v4 and v4 Turbo on 28 September 2026, after those results, so the score is for the earlier Eleven v3. Deepgram wasn't in the results.
- Price per 1,000 characters: ElevenLabs bills its API in dollars per character. Deepgram's rates are its pay as you go rates, and Aura-2 falls to $0.027 and Aura-1 to $0.0135 on a Growth plan, while Flux TTS stays at $0.045.
- Free usage: ElevenLabs' Free plan allows non-commercial use only. Deepgram's credit has no expiry and needs no card.
- First audio: The time from sending text to hearing the first sound. Each figure is the company's own claim, and the two use different measures.
- Delivery controls: Deepgram says it strips SSML, the markup language for pauses and pronunciation, and keeps synthesizing.
- Phone audio: Mu-law and A-law are the 8-bit formats phone networks carry.
- Security: A BAA is the business associate agreement a US healthcare customer signs with a vendor that handles patient data.
What does ElevenLabs do well, and where does it fall short for a voice agent?
ElevenLabs sells Eleven v4 and Eleven v4 Turbo, released on 28 September 2026, alongside Eleven v3, Multilingual v2, and Flash v2.5. Eleven v4 Turbo is the real-time model, with a median latency of about 100ms that ElevenLabs says excludes application and network latency, and ElevenLabs states a median time to first speech of about 150ms for v4 Turbo. Flash v2.5 is about 100ms in one place and about 75ms in another, and v3 Conversational is about 165ms or about 280ms, so ElevenLabs gives two figures for each.
The newest models change the streaming route. ElevenLabs says its standard Text to Speech WebSocket does not support eleven_v3 or eleven_v4 and routes v4 Turbo through Text to Dialogue. Delivery is by audio tags, and v4 and v3 take no SSML break tags, so pauses come from punctuation and tags. ElevenLabs offers BAAs to qualifying healthcare enterprises, and says text to speech can run inside a customer's own cloud.
Pros
- Audio tags for whispering, shouting, and laughing on v4 and v3
- Cloning from a $6 Starter plan, with 90+ languages on v4
- `ulaw_8000`, `alaw_8000`, and `pcm_8000` output for phone lines
- Private deployment in a customer's own cloud
Cons
- $0.08 per 1,000 characters on v4, and $0.04 on v4 Turbo
- The standard WebSocket doesn't take v3 or v4
- Professional cloning needs the $22 Creator plan
- Free plan limited to non-commercial use
Use ElevenLabs when:
- A team will direct delivery with audio tags and write pauses through punctuation
- A team will manage a plan tier for cloning access
- A team will route v4 Turbo streaming through the Text to Dialogue WebSocket
What does Deepgram Aura do well, and where does it fall short?
Deepgram sells three text to speech generations: Flux TTS on /v2/speak, Aura-2 on /v1/speak with voices in seven languages, and the English-only first-generation Aura. Flux TTS is conversational by design, and Deepgram says it "streams in as low as 80ms" and keeps tone across turns because conversational state persists. Flux speaks English plus 10 languages with built-in text normalization, and Deepgram says 40+ more run without it, and Deepgram marks 32 of 42 Flux TTS language codes as Early Access.
Delivery control is thinner. Deepgram's Flux TTS has a beta dial from -2 (calm) to 2 (animated), a speed setting from 0.5 to 1.5, inline IPA pronunciation in Early Access, and an inline pause on batch requests only, and it strips SSML with an INPUT_MARKUP_STRIPPED warning. Deepgram says Flux TTS can be self-hosted in a customer's own cloud, private cloud, or on-premise infrastructure, and offers a dedicated EU endpoint.
Pros
- A stated latency of as low as 80ms for Flux TTS at 24 kHz under production load
- $200 in credit with no card and no expiry
- Mu-law and A-law at 8 kHz, with a guide for Twilio Media Streams
- Self-hosted deployment and an EU endpoint, with SOC 2 Type 2 and BAAs for Enterprise customers
Cons
- Aura-2 is previous-generation, and current Flux TTS costs 50% more at $0.045 per 1,000 characters
- SSML is stripped with a warning
- 32 of 42 Flux TTS language codes are Early Access
- Inline pauses on batch requests only
Deepgram pricing
Deepgram bills per 1,000 characters of input text: Flux TTS $0.045, Aura-2 $0.030 on pay as you go and $0.027 on Growth, and Aura-1 $0.015 and $0.0135. Growth is prepaid credit from $4,000 a year with savings of up to 20%, and past the credits Deepgram bills the Growth rate plus 10%, weekly. A new account gets $200 in credit with no card and no expiry. Through 31 December 2026 Deepgram matches Flux TTS spend with $1 in credits for every $1 spent, up to $500.
Use Deepgram when:
- A team will pair text to speech with Deepgram speech recognition under one project
- A team will self-host Flux TTS or send traffic to the EU endpoint
- A team will steer delivery with a calm-to-animated dial and a speed setting
Which sounds more human, ElevenLabs or Deepgram Aura?
ElevenLabs' Eleven v3 ranked 12th of 13 AI voices at 938 in Design Arena's Audio Realism Bench results of 4 August 2026, and Deepgram wasn't in those results. Elo is a rating that rises each time listeners pick a voice over another, and Design Arena's Audio Realism Bench article describes native speakers choosing which of two recordings of the same line sounds more like a real person.
- #1BlandBland Speech v31365
- 2ndMicrosoftMAI-Voice-21224
- 3rdxAIGrok TTS1156
- 4thOpenAIGPT Realtime 21119
- 5thGoogleGemini 2.5 Pro TTS1109
- 12thElevenLabsEleven v3938
Ranks are among 13 AI voices, and ElevenLabs'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.
That leaves Deepgram without a score from the test and ElevenLabs with a score for a model it has since replaced. ElevenLabs released v4 and v4 Turbo on 28 September 2026, after the results. A buyer picking on sound alone would audition both. Bland Speech v3 ranked 1st at 1365, and the human recordings in the same pool scored 1500.
Why do people comparing ElevenLabs and Deepgram choose Bland Speech?
People comparing ElevenLabs and Deepgram choose Bland Speech when the buyer wants a voice with a first-place score in the 4 August 2026 results and one standard rate with no plan tier: Bland Speech costs $0.04 per 1,000 characters, which is $40 per million, with 50,000 characters free once. That equals ElevenLabs' v4 Turbo rate and sits below Flux TTS at $45 per million, and above Aura-2 at $30 and Aura-1 at $15, so a team reading rate sheets alone would pick Aura.
Deepgram's 80ms and ElevenLabs' median of about 100ms are both lower than the 199ms Bland publishes to first audio, and each company defines its own measure. Bland Speech v3 ranked 1st in the results at 1365, 141 points above the next voice, and Bland Speech streams model tokens over a WebSocket at WSS /v2/tts/ws, where a new context interrupts the turn that is playing. Output is mu-law at 8 kHz or PCM from 8 to 48 kHz. Bland Speech reads performance tags in square brackets, such as [laughs] or [say warmly], and takes expressiveness and stability settings from 0.0 to 1.0.
Bland is SOC 2 Type II audited and HIPAA compliant with a signed BAA, and FedRAMP 20x Class C certified at Moderate, granted 17 November 2026. Enterprise customers get dedicated orchestration and dedicated GPUs run by Bland. Enterprise pricing is custom, set with Bland's sales team. The text to speech API page has the endpoints and the Bland Speech page has the studio. Bland Speech vs Deepgram Aura sets Bland against Deepgram, the Deepgram Aura review and Deepgram Aura alternatives cover Deepgram on its own, and voice agent text to speech compared adds Cartesia.
Pros
- Ranked 1st of 13 AI voices on Audio Realism Bench on 4 August 2026
- One standard rate of $0.04 per 1,000 characters with no plan tier
- Performance tags and expressiveness and stability controls
- WebSocket streaming with barge-in and mu-law output at 8 kHz
Cons
- More than a hobby project needs, because Bland Speech is built for production calls
This pair splits price against expression, and neither company's current model has a listening score to set beside the other. For phone calls and voice agents, Bland Speech is the one that has a score, at the same rate as ElevenLabs' v4 Turbo.
Frequently asked questions
Is Deepgram Aura still current?
Aura-2 is previous-generation and the first-generation Aura is English only, according to Deepgram. Deepgram's current text to speech model is Flux TTS, which Deepgram recommends for all new builds.
Is ElevenLabs or Deepgram cheaper?
Deepgram's Aura-1 and Aura-2 are cheaper at $0.015 and $0.030 per 1,000 characters against $0.04 for ElevenLabs v4 Turbo, and Flux TTS is dearer at $0.045. ElevenLabs v4 costs $0.08. Bland Speech costs $0.04.
Which has lower latency, ElevenLabs or Deepgram?
Deepgram claims as low as 80ms for Flux TTS at full 24 kHz fidelity under production load, and ElevenLabs claims a median of about 100ms for v4 Turbo excluding application and network latency. Each figure is the company's own claim. Bland publishes 199ms to first audio.
Can Deepgram or ElevenLabs run in a customer's own cloud?
Both describe it. Deepgram says Flux TTS supports self-hosted deployment in a customer's own cloud, private cloud, or on-premise infrastructure, and ElevenLabs says text to speech can run inside a customer's own cloud infrastructure.
Do ElevenLabs and Deepgram work with Twilio, Pipecat, and LiveKit?
Both do. Deepgram has a Twilio guide that requests mu-law at 8 kHz to match Twilio's Media Streams format, and Pipecat and LiveKit each support both companies as text to speech services. ElevenLabs has a guide for sending generated audio through Twilio. Bland Speech has a first-party Pipecat service and a LiveKit Agents plugin in Python.
Are ElevenLabs and Deepgram HIPAA compliant?
ElevenLabs offers BAAs to qualifying healthcare enterprises, and Deepgram says it is HIPAA compliant and signs BAAs for Enterprise customers handling electronic protected health information. Bland is HIPAA compliant with a signed BAA.
