Fish Audio Review (2026)
Updated October 2, 2026
Short version: Fish Audio is the value buy in AI voice. Plus is $15 a month for up to 200 minutes, the API is pay-as-you-go with no minimum, the community library is enormous and the model weights are published. Where it is thin is the creator workflow: per-generation character caps, a Story Studio we could not evaluate, no subtitle export that we could find, and language counts that change from page to page. We make a competing product, so this page sticks to what Fish Audio publishes.
What Fish Audio is
Fish Audio is a voice platform built around its own models, currently led by S2.1 Pro, with a web app, a community voice library its home page puts at more than two million voices, and a developer API. The Fish Speech model weights are published on GitHub under the Fish Audio Research License.
What Fish Audio does best
- Price per minute. Plus is about $0.075 a minute, or $0.055 billed yearly; Pro is about $0.06. See Fish Audio pricing explained.
- A real API. Pay-as-you-go at $15 per million UTF-8 bytes on S2.1 Pro, with no subscription fee or monthly minimum, according to its documentation.
- Voice choice. The largest library of any tool we compare, mostly community-made, plus instant cloning from about 10 seconds of audio.
- Direction in plain language. On the S2 models, bracketed tags are free-form natural language rather than a fixed list.
- Open weights. Teams that want to run or study the model can, within the research license.
Where it limits you
- Per-generation caps. 500 characters on Free, 15,000 on Plus, 30,000 on Pro. A chapter is often more than that.
- Creator workflow. A Story Studio product is listed, but its page would not load for us without an account, so we cannot say what it does. We found no subtitle export on the pages we read.
- Language counts move. The home page says 30+, the documentation says 83 for S2.1 Pro, and the cloning page says zero-shot cloning in 13 languages.
- Free is non-commercial, and unused minutes do not roll over on any plan.
- The license is a research license. Open weights are not the same as a permissive license; read it before building on them.
Fish Audio facts are from fish.audio, fish.audio/plan, fish.audio/voice-clone, fish.audio/terms, docs.fish.audio and the fish-speech repository on GitHub as read on October 2, 2026 and may have changed.
Who should use Fish Audio
| Your work | Our call |
|---|---|
| A product that needs text to speech through an API | Fish Audio. Pay-as-you-go with no minimum. |
| Hundreds of minutes a month where price decides | Fish Audio. The cheapest minutes on our pages. |
| A specific character or accent from a community library | Fish Audio. |
| Books, series and long scripts with chapters, re-takes and subtitles | Compare. The workflow is where Fish Audio is thin. |
| Commercial publishing at low volume | Compare. The first commercial plan is $15. |
If your work is long-form narration
LeapFun is built for the last two rows: scripts live in a Studio project as chapters and paragraphs, a late edit is a one-paragraph re-take, every export is loudness-mastered with subtitle files, voices read 130+ languages on Leap Ultra, and commercial rights start at $6 a month. The trade-offs: a smaller voice library, no video dubbing, no streaming API.
These samples are unedited LeapFun output. The tags in brackets are read as direction, not spoken.
Previews above are real LeapFun output. Hit Generate to hear your script — 3,000 free credits on sign-up, no card required.
Frequently asked questions
Is Fish Audio worth it?
For an API or for high volume, yes: it is among the cheapest minutes available and the API has no monthly minimum. For long-form creator work, its per-generation caps and thin workflow are the trade-off.
Is Fish Audio free?
There is a free plan with 8,000 credits a month, up to 7 minutes and 500 characters per generation. Fish Audio's plan FAQ limits it to personal, non-commercial projects.
How many languages does Fish Audio support?
It depends which page you read: the home page says 30+, the documentation lists 83 for the S2.1 Pro model, and the cloning page says zero-shot cloning in 13 languages, as of October 2, 2026.
Is Fish Audio open source?
Its Fish Speech model weights are published on GitHub under the Fish Audio Research License, which is a research license rather than a permissive one. Read it before building on the weights.
What are the main drawbacks of Fish Audio?
Per-generation character caps, a creator workflow we could not evaluate, no subtitle export that we found, non-commercial free use, no rollover, and language counts that differ between its pages.
What is a good Fish Audio alternative for audiobooks?
LeapFun: a chapter-based Studio, paragraph re-takes, pronunciation rules, mastered exports with subtitles and commercial rights from $6 a month. It has a REST API too, but costs more per minute.
Try it on your script
3,000 free credits on sign-up, no card — generate a first take in seconds.
Start free See pricing