Head-to-head comparisons ยท Fish Audio vs Kokoro

Fish Audio vs Kokoro

Kokoro is free, small and fast, with voices you cannot change. Fish Audio is cheap, hosted and clones anyone in fifteen seconds. The choice is about whose voice you need.

Prices and policies as published by Fish Audio and Kokoro in October 2026. They change; the links go to the source. Hourly figures assume about 1,000 characters per finished minute.

Fish Audio and Kokoro side by side

Kokoro is a model, not a company, so several rows describe what running it yourself means. The third column is this site's own Studio, so you can see where it lands on the same rows.

  Fish Audio Kokoro Text to Speech Studio (this site)
Bills by Credits in the app, roughly 600 to 625 a minute; UTF-8 bytes on the API Nothing; the weights are free under Apache 2.0 Minutes of finished audio
Cheapest paid entry Plus $15 a month ($5.50 a month in the first year on the current offer) for 250,000 credits, up to 200 minutes Your own hardware, or a hosted API at under $1 per million characters $49 a month for 500 minutes with the API; $99 one-time for 300 minutes a month with the API
Free tier 8,000 credits a month, up to 7 minutes, personal use only Everything, including commercial use Free account, no card; a no-login free text to speech page
What a finished hour costs About $4.50 an hour on Plus at list price, $3.70 on Pro; the API is $15 per million bytes, about $0.90 an hour Under $0.06 an hour hosted; the electricity bill if you self-host $5.88 an hour on the $49 monthly plan; the $99 one-time plan starts near $20 an hour and falls every month you keep using it
Voices and languages A large community voice library; the site names eight core languages and claims 30+ 54 voices across 9 language variants in v1.0 354 library voices in 18 languages, a voice designer, celebrity-style voices
Cloning From 10 to 15 seconds of audio on every plan None; it is a fixed-voice model From one sample of up to 30 seconds; directions work on clones
Delivery control Emotion and delivery tags written into the text Speed only; no emotion or direction Bracket directions per line: [whisper], [seductive], [sigh], [laugh]
Adult content The terms prohibit submissions that contain sexually explicit content or pornography No policy; a model file has no terms of use Allowed within the AI usage policy; nothing involving minors, no real person's voice without consent
API Pay as you go; concurrency rises with lifetime spend Whatever host you run it on On the monthly plans and the $99 one-time tier
Built for Cost-driven creators and developers who want quick cloning and an open-weights lineage Developers who want a small, fast, free narrator and can live with a fixed voice set Creators who need emotion, cloning, a voice changer, audiobooks and adult content in one place

Hour figures divide the plan price by the minutes it includes, or multiply the per-character rate by roughly 60,000 characters an hour. Studio prices are the ones the Studio page shows today; the Studio's content rules are in its AI usage policy.

What Kokoro is

Kokoro is an open-weight text to speech model with 82 million parameters, released under the Apache 2.0 licence. Version 1.0 shipped in January 2025 with 54 voices across nine language variants, among them American and British English, Spanish, French, Hindi, Italian, Japanese, Brazilian Portuguese and Mandarin. It is small enough to run on a laptop CPU, and hosted APIs offer it for under $1 per million characters, under $0.06 an hour. There is no cloning: you pick from the 54 voices and adjust speed.

What Fish Audio is

Fish Audio is a hosted service built on its own open-weight lineage, with a community voice library, cloning from 10 to 15 seconds of audio on every plan, and emotion tags in the text. Its free plan gives 8,000 credits a month, up to 7 minutes, for personal use; Plus is $15 a month at list price for up to 200 minutes, about $4.50 an hour, with a first-year offer at $5.50; the API is $15 per million UTF-8 bytes, about $0.90 an hour in English.

Price and effort

Kokoro wins on price by definition: the model is free and a hosted call costs cents an hour. What you pay instead is set-up, hosting and the absence of anyone to call when it breaks. Fish Audio costs real money but is a product: an editor, an account, an API with rate limits that rise as you spend, and a company responsible for uptime.

Content rules

A model file has no terms of use; what you generate with Kokoro is governed by the law where you run it and the policy of any host you use. Fish Audio's terms prohibit submissions that contain sexually explicit content or pornography. For adult work, self-hosting Kokoro is allowed; using Fish Audio is not.

Who should pick which

Kokoro suits a developer who needs a dependable narrator in one of its languages, is happy with a fixed voice, and wants zero running cost or full control over the stack. Fish Audio suits a creator who needs their own voice or a specific character, wants emotion tags, and would rather pay a few dollars a month than maintain a server.

Where the Studio fits between them

The Studio is a hosted product like Fish Audio, with the things Kokoro cannot do: a clone from one 30-second sample, a voice designer, 354 library voices in 18 languages, and bracket directions such as [whisper] and [laugh] that work on clones. It bills by the minute, $49 a month for 500 minutes at $5.88 an hour, or $99 one-time for 300 minutes a month.

Against Fish Audio its difference is the policy: adult content is allowed within the usage policy, with the same hard lines on minors and consent. Against Kokoro its difference is that someone is on the other end.

Open the Text to Speech Studio

Fish Audio vs Kokoro: the questions people ask

Is Kokoro really free?

Yes. The weights are Apache 2.0 licensed, including commercial use. Running it costs hardware, or a hosted API at under $1 per million characters, under $0.06 an hour.

Can Kokoro clone a voice?

No. It ships 54 fixed voices across nine language variants and a speed control. Fish Audio clones from 10 to 15 seconds of audio on every plan.

Does Fish Audio allow NSFW?

No. Its terms prohibit submissions that contain sexually explicit content or pornography. A self-hosted Kokoro has no such terms.

Which sounds better?

Fish Audio's current models are larger and carry emotion tags; Kokoro's 82 million parameters buy speed and size, not range. For a steady narrator Kokoro is fine; for a performance, Fish Audio.