Fish Audio bills text to speech by the byte and ElevenLabs bills by the character, and the two units match only while a script is plain English. A byte is the unit computers store text in: an English letter takes 1 byte in UTF-8, the standard encoding, a character in Arabic takes 2, and most Japanese or Chinese characters take 3.
Fish Audio charges $15 per million UTF-8 bytes for s2.1-pro, its production model, and ElevenLabs charges $0.08 per 1,000 characters for Eleven v4, which is $80 per million. A team shopping on those two numbers should also weigh Bland Speech, which costs $40 per million characters and ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, where ElevenLabs' Eleven v3 ranked 12th.
Which is better, ElevenLabs or Fish Audio?
Fish Audio is the better pick for a team that puts price and open weights first, and ElevenLabs is the better pick for a team that wants a completed SOC 2 Type 2 audit and 90+ languages, as of November 2026. On English text Fish Audio's s2.1-pro costs about $15 per million characters, against $40 for ElevenLabs' Eleven v4 Turbo and $80 for Eleven v4.
ElevenLabs fits when
- A team will manage plan tiers and included-character allowances
- A brand needs professional clones verified by a voice-captcha step, from the $22 Creator plan
- A security team wants a vendor that holds SOC 2 Type 2, ISO 27001:2022, and PCI DSS 4.0.1 Level 1
Fish Audio fits when
- A team will forecast spend in UTF-8 bytes and set the model header deliberately, because a missing header falls back to s2.1-pro
- A team will run an open model on its own hardware under a community license that is free below USD 500,000 of annual revenue
- An enterprise will pay a $10K platform fee and a $10K monthly minimum on a 12-month commitment to run Fish Audio in a VPC, on-premise, or a sovereign cloud
What these rows mean
- Audio Realism Bench, 4 August 2026: Design Arena's blind listening test, where listeners hear two clips of the same line and pick the one that sounds more like a real person. A higher score means listeners picked that voice more often. Fish Audio wasn't in those results, and ElevenLabs' Eleven v4 models were released after them.
- Price: Bland Speech and ElevenLabs bill per character, and Fish Audio bills per UTF-8 byte. Fish Audio says 1 million bytes is about 180,000 English words, or about 12 hours of speech. Fish Audio's Drama 3 is $10 per million bytes through 30 November 2026, as a launch promotion.
- Free usage: ElevenLabs' Free plan allows non-commercial use only. Fish Audio says its free model comes with no first-audio or data-processing-agreement guarantees.
- First audio: The pause between sending text and hearing the first sound, as each company publishes it. Fish Audio also says "<70" ms of latency for enterprise customers, unlabelled as time to first audio.
- Phone audio: Mu-law is the 8-bit format US phone networks and Twilio carry. Fish Audio's text to speech formats are mp3, wav, pcm, and opus, so audio for a mu-law line takes a conversion step.
What are the catches with ElevenLabs and Fish Audio?
ElevenLabs' catch is its $80 flagship rate and a standard WebSocket that skips Eleven v4, and Fish Audio's catch is billing by the byte, a free model without guarantees, and output formats limited to mp3, wav, pcm, and opus.
ElevenLabs
ElevenLabs sells Eleven v4, which speaks 90+ languages, and Eleven v4 Turbo, which ElevenLabs says has a median latency of about 100ms, both launched on 28 September 2026. Eleven v4 requests go through the Text to Dialogue API, because the standard text to speech WebSocket doesn't take them.
An instant clone takes 10 seconds of audio on Eleven v4 from the Starter plan. A professional clone takes about 30 minutes of audio from the Creator plan up, and the voice owner reads a captcha aloud to confirm the voice is theirs. ElevenLabs says enterprises can run text to speech inside their own cloud infrastructure.
Pros
- 90+ languages across the Eleven v4 family
- Professional clones with voice-captcha verification
- Mu-law and A-law output at 8 kHz for phone lines
- SOC 2 Type 2, HIPAA, and PCI DSS 4.0.1 Level 1 on its trust center, with BAAs on Enterprise
Cons
- Eleven v3 ranked 12th of 13 AI voices on Audio Realism Bench on 4 August 2026
- $80 per million characters for Eleven v4 on the API
- The standard WebSocket doesn't take Eleven v3 or Eleven v4
- Characters sold through monthly plans, with a non-commercial Free plan
Fish Audio
Fish Audio is a voice AI platform from Hanabi AI Inc., with a web app, a REST API, and SDKs, and it announced $52 million of seed funding on its first anniversary. Fish Audio's production model is s2.1-pro, supporting 83 languages. Drama 3, launched on 13 October 2026, takes plain-language direction for a delivery. S2 Pro, the previous generation, was open-sourced on 9 March 2026, and s1 retires on 31 December 2026.
Text can stream in over a WebSocket word by word, or go over HTTP streaming. Inline bracket tags steer delivery in natural language rather than from a fixed set, such as [whispers sweetly], and one request can carry several speakers. Professional clones run only through the web app, need at least 10 minutes of audio, and come with 1 slot on the Plus plan and 5 on Pro, after the speaker confirms permission.
On security, Fish Audio says a SOC 2 audit is underway with the report available to customers under NDA once complete, offers HIPAA-aligned configurations and a BAA for qualifying healthcare workloads, and offers zero data retention on enterprise contracts. Data stays in the United States by default.
Pros
- $15 per million UTF-8 bytes for s2.1-pro, with no subscription for API access
- 83 languages on s2.1-pro
- Open S2 weights, and an enterprise tier that runs in a customer's VPC, on-premise, or sovereign cloud
- Natural-language direction tags and multi-speaker dialogue
Cons
- Billed per UTF-8 byte, so Japanese or Arabic text costs more per character than English
- The free model carries no first-audio or data-processing-agreement guarantees
- Output formats are mp3, wav, pcm, and opus, so a mu-law phone line needs a conversion step
- A SOC 2 audit still underway
- Professional clones only through the web app
What do ElevenLabs, Fish Audio, and Bland Speech cost?
Fish Audio is the cheapest of the three on English text at $15 per million bytes, Bland Speech sits in the middle at $40 per million characters with no monthly plan, and ElevenLabs charges $40 to $80 depending on the model.
ElevenLabs pricing
ElevenLabs charges $0.08 per 1,000 characters for Eleven v4 and $0.04 for Eleven v4 Turbo on its API, and sells the characters in monthly plans: Starter at $6, Creator at $22, Pro at $99, Scale at $299, and Business at $990, with Enterprise custom. At the Eleven v4 rate those plans include 75,000, 275,000, 1,238,000, 3,738,000, and 12,375,000 characters, and the Free plan includes 10,000 for non-commercial use.
Fish Audio pricing
Fish Audio's API has no subscription fees or monthly minimums. Text to speech costs $15 per million UTF-8 bytes on s2.1-pro and s2-pro, $50 on Drama 3, and $0 on the free s2.1-pro-free model. Concurrency follows prepaid spend: 5 requests at once under $100 paid, 15 from $100, and 50 from $1,000. Fish Audio's web app plans are priced in pounds, from a Free plan with 8,000 credits a month to Pro at £76.99 a month.
Bland Speech pricing
Bland Speech costs $0.04 per 1,000 characters, directed or not, in the studio and the API, pay as you go with no plan. A new account gets 50,000 characters free, once. Enterprise pricing is custom, set with Bland's sales team.
What a month of speech costs on each
For English text, where a byte is a character, 100,000 characters a month costs $4 on Bland Speech, $1.50 on Fish Audio's s2.1-pro, and $22 on ElevenLabs' Creator plan for Eleven v4 or $6 on Starter for Eleven v4 Turbo. At 1 million characters, the figures are $40, $15, and $99 on Pro. At 10 million, they are $400, $150, and $990 on Business, assuming usage stays inside ElevenLabs' included characters. A million Japanese characters is about 3 million bytes, which is about $45 on Fish Audio's s2.1-pro and more than Bland Speech's $40.
Which sounds more human, ElevenLabs, Fish Audio, or Bland Speech?
In Design Arena's Audio Realism Bench results of 4 August 2026, Bland Speech v3 ranked first of 13 AI voices at 1365, ElevenLabs' Eleven v3 ranked 12th at 938, and a human recording scored 1500. Fish Audio wasn't in those results.
- #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.
Listeners hear two clips of the same line without knowing which model made which, and pick the one that sounds more like a real person. Scores are Elo ratings, the chess-style number that rises every time a voice is picked over another. Fish Audio launched Drama 3 on 13 October 2026 and ElevenLabs launched Eleven v4 on 28 September 2026, both after those results. Design Arena set out its method, with one female and one male American English voice per model, in an article on Audio Realism Bench.
Why do people comparing ElevenLabs and Fish Audio choose Bland Speech?
People comparing ElevenLabs and Fish Audio choose Bland Speech for a phone line because the voice ranked first among AI voices on 4 August 2026, the audio arrives as mu-law, and the rate counts characters rather than bytes. Bland publishes 199ms to first audio, slower than the 100ms ElevenLabs and Fish Audio publish, and Bland Speech costs more than Fish Audio's s2.1-pro.
A WebSocket at /v2/tts/ws takes text as a language model writes it, and a new context interrupts the turn that is playing. Output is mu-law at 8 kHz or PCM from 8 to 48 kHz, so audio goes onto a call without conversion. An endpoint that follows the OpenAI speech contract lets a team move over by changing a base URL, a key, and a model name. The text to speech API page has the endpoints.
Bland is SOC 2 Type II audited, HIPAA compliant with a signed BAA, and FedRAMP 20x Class C certified at Moderate, granted on 17 November 2026. Bland Speech is a closed-source model Bland built itself.
Pros
- Bland Speech v3 ranked first of 13 AI voices on Audio Realism Bench on 4 August 2026
- One standard rate, $0.04 per 1,000 characters, counted in characters rather than bytes, with no monthly plan
- Mu-law and PCM output ready for phone calls
- Instant clones from about 10 seconds, private to the customer's organization
Cons
- Built around phone calls, which is more than a one-off narration needs
Bland has a page on Bland Speech against Fish Audio, a Fish Audio review, and a list of Fish Audio alternatives. The Bland Speech page has the studio and the price, the text to speech comparison hub covers every vendor, and the rival comparisons cover other pairs.
Frequently asked questions
Is Fish Audio cheaper than ElevenLabs?
On English text, yes. Fish Audio charges $15 per million UTF-8 bytes for s2.1-pro, where an English letter is 1 byte, against $80 per million characters for ElevenLabs' Eleven v4 and $40 for Eleven v4 Turbo. Text in scripts that use 2 or 3 bytes per character costs more on Fish Audio than the character count suggests.
Is ElevenLabs better than Fish Audio?
ElevenLabs is better for a team that needs a completed SOC 2 Type 2 audit, 90+ languages, and clones verified by voice captcha. Fish Audio is better for a team that wants a lower price, open weights, and an enterprise tier that runs in the customer's own environment.
Can Fish Audio run on a company's own servers?
Fish Audio says its enterprise tier runs in a VPC, on-premise, air-gapped, or in a sovereign cloud, from a $10K one-time platform fee and a $10K monthly minimum on a 12-month commitment. Fish Audio's open source license is a community license, free for research and non-commercial use, and free for commercial use only below USD 500,000 of annual revenue.
Do ElevenLabs and Fish Audio support phone audio?
ElevenLabs returns mu-law and A-law at 8 kHz, the formats phone networks carry. Fish Audio's text to speech formats are mp3, wav, pcm, and opus, with a default of 44,100 Hz, so audio for a mu-law phone line takes a conversion step. Bland Speech returns mu-law at 8 kHz directly.
Can ElevenLabs and Fish Audio clone a voice from a short recording?
Both clone from short clips. ElevenLabs makes an instant clone from 10 seconds on Eleven v4, and Fish Audio asks for at least 10 seconds per clip, with a minute or two improving fidelity. Bland Speech makes an instant clone from one clean sample of about 10 seconds.
Which text to speech API suits a voice agent on a phone line?
Bland Speech suits a voice agent on a phone line, because Bland Speech v3 ranked first of 13 AI voices in the Audio Realism Bench results of 4 August 2026 and returns mu-law at 8 kHz. Fish Audio suits a team that puts price and self-hosting first, and ElevenLabs suits a team that needs language breadth.
