whatmodel.sh / tts / Fish Audio Drama 3

Fish Audio Drama 3

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Fish Audio Drama 3 is an open-weights text-to-speech model from Fish Audio, released October 2026. Its published size is 35B parameters (3B active per token). Licence: fish-audio-research-license.

Catalogue identity notes

Public checkpoint with maker-documented runtime. Memory is estimated from parameters and precision, not measured; no API price verified in this pass.

Model information sources

Release date ↗

48.5±12.7#71 of 119 text-to-speech modelsfrom 1 benchmark result across 1 source* thinly tested, treat the rank as provisional

Score within the Vapi Humanness · Bradley–Terry rating, 2026-09-25 snapshot comparison set only.

$ how it compares

The text-to-speech models ranked either side of Fish Audio Drama 3, out of 149 with a score, then more from Fish Audio.

#modelscorewhereRAMreleased
87Parler-TTS Mini v0.1 Hugging Face (Parler-TTS)49±15local~2.6GB2024-04
88Fish Speech 1.2 Fish Audio49±15local~1.2GB2024-07
89Fish Audio Drama 3 Fish Audio49±13local~85GB2026-10
90SILMA TTS v1 SILMA AI48±10local~2.7GB2026-03
91FireRedTTS (v1) FireRedTeam (Xiaohongshu)48±3local~3.1GB2024-09
more from Fish Audio
4OpenAudio S1 Fish Audio71±8api $15.00/M chars2025-06
32Fish Audio S2.1 Pro Fish Audio63±3api $15.00/M chars2026-06
36Fish Audio S2 Pro Fish Audio63±1api $15.00/M chars +local~11GB2026-03

see all text-to-speech models · find the ones that fit your machine

$ what it's good at

other tests1 not in a task · 35% of its score
testresultvs the others in its rundifficultycountswhy
MiniMax multilingual 24-language mean error (Drama 3 report) lab3.1% ✎█·█········█ beats 1/2avg 2.9%▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 35%blunt · small run

$ where to run it

Other ways to run it

  1. Run it as published · needs ~84.5 GB, NVIDIA only · from huggingface.co/fishaudio (official)
    sgl-omni serve --model-path fishaudio/drama-3 --port 8000

Official weights and documentation: huggingface.co/fishaudio ↗

$ how the score was worked out · Fish Audio Drama 3 technical… 35%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
Fish Audio Drama 3 technical report · lab-reported1CC BY 4.0
prior · pull toward the middle
weighted average= 48.5

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json