Fish Audio just made their best text to speech model free to call as an API. Not a trimmed down version, not a trial. S2.1 Pro is the same state of the art voice model paying customers get, and any developer can now hit it with no hard usage cap.

Nobody else's free tier is close

This is the part that makes it worth paying attention to. Here is what the free tiers actually look like across the major providers, from Fish Audio's own comparison against public pricing pages in June 2026.

Fish Audio S2.1 Pro Free gives you unlimited usage subject to a fair use policy, the latest model, 83 languages, and voice cloning included.

ElevenLabs free tier caps you at 10,000 credits a month, does not give you the latest model, and limits voice cloning.

OpenAI TTS has no free quota at all. TTS-1 and TTS-1-HD cover 57 languages and there is no voice cloning.

Google Cloud TTS has a 4 million character legacy free quota, but Gemini TTS is paid only, and again no voice cloning.

So every other option either gates the good model behind a paywall or hands you a few minutes and stops. Fish is handing over the flagship.

The specs are actually production grade

Free is only interesting if the model is good enough to ship with, and this one is.

83 languages live in a single model, so you are not stitching together per language endpoints. Time to first audio sits around 90 milliseconds, which is fast enough for live voice agents where anything above a couple hundred milliseconds starts to feel like a delay. And voice cloning from a reference sample is on the free tier, which is normally the first thing providers lock up.

Switching to it is one line

If you are already calling Fish Audio, you do not rewrite anything. Grab an API key and set the model header to s2.1-pro-free. Same endpoint as the paid API, same request shape.

The catch worth knowing

There is one, and it is timing rather than fine print. The free window currently runs through August 31, 2026. Fish has already extended it once, so it may well move again, but if you have been meaning to test a voice model properly, this is a genuinely free window to do it in rather than burning credits to find out whether the quality holds up.

Full details are on Fish Audio's announcement post.