A listening test scores whichever model it was given, and that isn't always the model a company leads with. Design Arena's Audio Realism Bench results of 4 August 2026 scored Inworld TTS-1.5 Max and MiniMax Speech-02 HD, while Inworld now leads with Realtime TTS-2, launched 5 May 2026, and MiniMax with its Speech 2.8 family, released 23 January 2026. Fish Audio wasn't in those results at all.
Past that test, a buyer comparing these four is mostly comparing each company's published claims. Bland's text to speech comparison hub covers more providers.
Which is best for a phone call: Bland Speech, Inworld, MiniMax, or Fish Audio?
For a phone call, Bland Speech is the pick of the four as of November 2026: Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, one standard rate applies with no plan, and the audio comes back as 8 kHz mu-law, a telephony format. Inworld and Fish Audio publish lower rates per million characters or bytes and Inworld a smaller first-audio figure, so a team choosing on price or speed alone has reasons to look past Bland Speech.
What these rows mean
- Audio Realism Bench: Design Arena's blind listening test, where a listener hears two clips of the same line from two models and picks the one that sounds more like a real person. The votes become an Elo rating, a score that rises each time a voice is picked.
- Price per million characters: Each company's published rate with the model named. Fish Audio bills per million UTF-8 bytes, which equals characters for plain English text and runs higher for other scripts.
- Phone audio: Mu-law is 8-bit compressed audio at 8 kHz, used on North American and Japanese phone lines and by Twilio. A-law is the other common phone encoding.
How fast does each one start talking?
Time to first audio is the gap between sending text and receiving the first sound, and a voice agent's reply can't begin before it. Bland publishes 199ms on a streamed request, counted to the first audio chunk. Inworld claims as low as 50ms for Realtime TTS-2 Flash and 90ms at the 90th percentile for Realtime TTS-2, measured on Coval, and separately states 100ms and 20ms counted on the server and excluding network time. Fish Audio says "<70 ms" for enterprise without naming what it counts, and MiniMax describes speech-2.6-hd as "ultra-low latency". On those numbers Bland Speech starts slower than Inworld and Fish Audio, and a team that needs the smallest figure should count that against it.
What does each send down a phone line?
Bland Speech, Inworld, and MiniMax each return mu-law at 8 kHz, the encoding North American phone lines use, and Inworld adds A-law. Fish Audio's formats are mp3, wav, pcm, and opus at 44.1 kHz by default, so a team building a phone agent on Fish Audio converts to 8 kHz mu-law itself before audio reaches a carrier.
Where do Inworld, MiniMax, and Fish Audio fall short next to Bland Speech?
Inworld steers delivery only on its slower model, MiniMax charges the highest rates of the four, and Fish Audio leaves conversion to phone audio to the team, while Bland Speech returns phone-ready audio at one standard rate.
Where Inworld falls short
Inworld sells Realtime TTS-2, launched on 5 May 2026, and Realtime TTS-2 Flash, launched on 9 August 2026, and aims its marketing at consumer-facing applications such as companions, games, and education. The catch is how the two models split: Inworld says steering, the plain-language direction of delivery, works only on TTS-2, so the fast Flash model ignores it. Inworld also says 500+ languages and dialects in one place and 200+ languages and locales in another, and gives two sets of latency figures for the same models, so a buyer should confirm the number that matters in a trial.
Pros
- $15 per million characters on TTS-2 Flash and $25 on TTS-2, falling to $12.50 on TTS-2 on the $1,500 Growth plan
- Instant clone from as little as 3 seconds of audio, included on every plan
- Mu-law and A-law output at 8 kHz, and an OpenAI-compatible endpoint
- SOC 2 Type II, with HIPAA and a BAA on Enterprise or as a Growth add-on
Cons
- TTS-1.5 Max ranked 13th of 13 AI voices on Audio Realism Bench on 4 August 2026, and TTS-2 Flash was launched after those results
- Steering works on TTS-2 only, not on the fast Flash model
- Zero data retention starts on Developer and doesn't cover clone samples
Use Inworld when:
- A team will choose between TTS-2 for steered delivery and TTS-2 Flash for speed, and accepts that steering stays off on Flash
- A team will move to a monthly plan to bring the per-million rate down from $25
Where MiniMax falls short
MiniMax's current speech models are speech-2.8-hd at $100 per million characters and speech-2.8-turbo at $60, the highest rates of the four against $40 for Bland Speech. A cloned voice is deleted if it goes unused for 7 days, and rapid cloning is billed at $1.5 per voice on first use.
Pros
- 40 languages on the 2.6 and 2.8 models
- 19 inline interjection tags, such as laughs, sighs, and breath, on the 2.8 models
- Voice design from a written description through the API
- A bidirectional WebSocket that takes LLM output at any granularity and supports a cancel message
- Mu-law output at 8 kHz
Cons
- $100 and $60 per million characters, the highest of the four
- Speech-02 HD ranked 7th of 13 AI voices on Audio Realism Bench on 4 August 2026, and Speech 2.8 wasn't in those results
- A cloned voice is deleted after 7 days without use
- Whisper isn't available on the 2.8 models
Use MiniMax when:
- A team scripts delivery with inline sound tags and will build for the 2.8 models to use them
- A team will design voices from text descriptions and pay $3 for each voice
Where Fish Audio falls short
Fish Audio names s2.1-pro as its recommended production model at $15 per million UTF-8 bytes, and drama-3, launched on 13 October 2026, as a directable model at $50, with a launch price of $10 through 30 November 2026. A free model, s2.1-pro-free, runs the same model at $0 under fair-use limits, and Fish says it carries no first-audio or data processing agreement guarantees. Professional voice cloning is available in the web app only, on a paid plan.
Pros
- $15 per million UTF-8 bytes on s2.1-pro, with no subscription for API access
- 83 languages on s2.1-pro
- Natural-language bracket tags, not limited to a fixed set, and multi-speaker dialogue in one request
- A professional clone confirms the speaker gave permission before the voice is released
- A free model for development, and a 99% uptime SLA on Enterprise
Cons
- The free model has no first-audio or data processing guarantees
- Professional cloning is in the web app only, on a paid plan
- A SOC 2 audit is underway, with no completed report yet
Use Fish Audio when:
- A team will convert mp3, wav, pcm, or opus output to 8 kHz telephone audio before it reaches a carrier
- A team will write bracket tags in plain language and test how drama-3 and s2.1-pro read them
What do Bland Speech, Inworld, MiniMax, and Fish Audio cost?
One million characters of speech costs $40 on Bland Speech, $15 or $25 at Inworld's On-Demand rates, $60 or $100 on MiniMax's API, and $15 or $50 at Fish Audio's list rates, so Bland Speech sits in the middle of the four on price.
Bland Speech costs $0.04 per 1,000 characters, which is $40 per million, pay as you go and the same in the studio and the API, with no monthly plan. A new account gets 50,000 characters free once, and credits don't expire. Some public-library voices add a per-character creator fee on top. Enterprise pricing is custom, set with Bland's sales team.
Inworld's On-Demand plan is $0 a month and includes up to 70 minutes of text to speech and a commercial license, and its paid plans from $25 to $1,500 a month each return their fee as credit while TTS-2 falls from $25 to $12.50 per million. MiniMax's audio packages run from $5 to $999 a month.
At 10 million characters a month, worked from per-use rates with no tax or phone charges, Bland Speech costs $400, Inworld $150 or $250, MiniMax $600 or $1,000, and Fish Audio $150 or $500 for English text. At that volume Inworld and Fish Audio come in at well under half of Bland Speech's bill on their lowest-priced models, and a team with steady volume and a model it likes can price that difference. The case for Bland Speech rests on the realism result and on phone-ready audio at one standard rate, with no plan tier or per-byte arithmetic to manage.
Which sounds more human, Bland Speech, Inworld, MiniMax, or Fish Audio?
Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, at 1365, where the human recording scored 1500. MiniMax's Speech-02 HD ranked 7th at 1093 and Inworld's TTS-1.5 Max ranked 13th at 763. Design Arena describes the top of its result in its own words: "Second only to human speech is Bland Speech v3 by Bland AI, with an Elo of 1365."
- #1BlandBland Speech v31365
- 2ndMicrosoftMAI-Voice-21224
- 3rdxAIGrok TTS1156
- 4thOpenAIGPT Realtime 21119
- 5thGoogleGemini 2.5 Pro TTS1109
- 7thMiniMaxSpeech-02 HD1093
Ranks are among 13 AI voices, and MiniMax'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.
The test used one female and one male American-English voice per model, with phone-agent prompts as one of three categories, so a team working in other languages or on long narration should listen for itself. Inworld's Realtime TTS-2, MiniMax's Speech 2.8, and Fish Audio's drama-3 weren't scored, so the 4 August placings say nothing about how those compare.
Why is Bland Speech the pick over Inworld, MiniMax, and Fish Audio for phone calls?
Bland Speech is the pick over Inworld, MiniMax, and Fish Audio when the product is a phone call, because Bland Speech v3 was built for calls, returns mu-law at 8 kHz, and bills one standard rate with no plan to size. Bland runs its own model and speech instead of wrapping a third party's.
Text goes to /v2/tts over HTTP, or streams into a WebSocket at /v2/tts/ws, where a new context interrupts the turn that is playing, and performance tags such as [laughs] or [say warmly] shape the delivery. The text to speech API page carries the endpoints, and Bland Speech has the studio.
Bland, the company behind the model, is SOC 2 Type II audited, HIPAA compliant with a signed BAA, and FedRAMP 20x Class C certified at Moderate, granted 17 November 2026, and its trust page says speech synthesis runs inside the FedRAMP boundary. Every clone is private to the customer's organization, and a professional clone goes through Bland's consent and rights review.
Pros
- Ranked first 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 and no credit meter
- Mu-law at 8 kHz and PCM from 8 to 48 kHz
- Performance tags, plus a WebSocket where a new context interrupts a turn
- A first-party Pipecat service, a LiveKit plugin, and an OpenAI-compatible endpoint
- An instant clone from about 10 seconds of audio
Cons
- More than a hobby project needs, because Bland Speech is built for production calls
The two-way versions are Bland Speech vs Inworld, Bland Speech vs MiniMax, and Bland Speech vs Fish Audio, and the rival pair is Inworld vs MiniMax. Every provider in the series is on the rival comparisons page.
Frequently asked questions
Is Bland Speech better than Inworld, MiniMax, or Fish Audio?
For phone calls, yes: Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, ahead of MiniMax's Speech-02 HD and Inworld's TTS-1.5 Max. Inworld and Fish Audio cost less per million, which matters more to a team that weighs price over the realism result.
Which of the four is cheapest per million characters?
Inworld's TTS-2 Flash and Fish Audio's s2.1-pro both cost $15 per million, ahead of $25 for Inworld's TTS-2, $40 for Bland Speech, $50 for Fish Audio's drama-3 at list price, and $60 to $100 for MiniMax. Fish Audio counts UTF-8 bytes, so text in other scripts costs more per character than English.
Can each one clone a voice?
Yes. Bland Speech needs about 10 seconds for an instant clone and 30 minutes or more for a professional one, Inworld as little as 3 seconds, and Fish Audio and MiniMax at least 10 seconds.
Do they work with Pipecat and LiveKit?
Bland Speech, Inworld, and Fish Audio each have a Pipecat service maintained by the Pipecat team, and all four have a LiveKit Agents plugin.
How do the four compare on security for regulated calls?
Bland is SOC 2 Type II audited and HIPAA compliant with a signed BAA, and adds FedRAMP 20x Class C certification at Moderate, granted 17 November 2026, with speech synthesis inside the FedRAMP boundary. Inworld says it is SOC 2 Type II certified and HIPAA and GDPR compliant, with a BAA on Enterprise or as a Growth add-on, and Fish Audio says a SOC 2 audit is underway.
