Fish Audio's S2.1 Pro: production TTS, free via API through Aug 31
The model ElevenLabs would charge for is now a free API call — 83 languages, ~90ms first audio, voice cloning from seconds, and a one-line migration.
Copy markdown
What just went free
S2.1 Pro — Fish Audio's flagship voice model, previously held back from open release — is now callable free via API (model id `s2.1-pro-free`) through Aug 31, 2026 under a Fair Use Policy with no hard character cap. There's no uptime SLA and requests may be used to improve the model, so it's built for prototyping, not production traffic.
The pricing wedge, in context
ElevenLabs' free tier taps out around 6–10 minutes of audio, OpenAI's TTS has no free tier, and Gemini TTS bills from the first token. Fish Audio is handing builders unmetered access to a production-grade voice model — the play is to get you switched before you ever pay.
83 languages, ~90ms to first audio
The model covers 83 languages — English, Japanese, Chinese, Korean, Arabic, Spanish and more — and returns first audio in about 90ms on the standard API, with a 2x+ throughput bump under high concurrency. That's fast enough for real-time voice agents and live dubbing, not just batch narration.
Voice cloning from a short sample
S2.1 Pro clones a voice from a few seconds of reference audio across all 83 languages and outputs MP3, WAV, or Opus. Trained on 10M+ hours of audio, it posts a 61% win rate over the previous S2 Pro in blind listening tests.
Migration is one line
Already piping text through another TTS? Point it at `s2.1-pro-free` — grab a key at fish.audio/app/api-keys and send a JSON body with your text and a reference_id. Products over $1M ARR are asked to contact Fish before leaning on the free tier.
The catch to plan around
Free access has already been pushed back twice (from a July 24 cutoff to Aug 31), so treat the date as soft but not permanent — and with no uptime guarantee, don't wire it into anything customer-facing yet. Paid plans exist for latency and uptime SLAs when you're ready to ship.
Build this weekend
A free, low-latency, 83-language voice API opens up projects that used to get metered into oblivion: a multilingual audiobook generator, a real-time voice for your CLI agent, or a one-pass dub over your demo videos. Until Sept 1, the only cost is the clock.