October 13, 2026
Drama 3 Preview: open voice model beats Eleven v4 on intelligibility, 93.2%
Fish Audio's open weights read text more accurately than the paid Eleven v4: 93.2% vs 91.7% in the Artificial Analysis test of Oct 13.
Artificial Analysis
@artificialanlys
Drama 3 Preview from Fish Audio is the new #1 open weights model in the Artificial Analysis text-to-speech leaderboard. It also tops our Pronunciation Robustness leaderboard, ahead of Eleven v4 and Eleven v4 Turbo. Drama 3 Preview is a new open weights text-to-speech model from @FishAudio. It was built for multi-character scenes and long-form text: it supports voice cloning and follows instructions in plain language. The weights are available on Hugging Face under the Fish Audio Community License. Key takeaways: ➤ Pronunciation Robustness: Drama 3 Preview scores 93.2% and ranks first, with Eleven v4 at 91.7% and Eleven v4 Turbo at 90.1%. The model is first in the Standalone Terms and Preserving Exact Sequences categories and second in Expanding Shorthand. ➤ Provider Voice: Drama 3 Preview has an Elo of 1243 (+16/-16) across 1,857 appearances, placing 10th and gaining 101 points over the earlier Fish Audio S2.1 Pro at 1142. ➤ Controlled Voice: Drama 3 Preview ranks 14th with an Elo of 1058 (+15/-15) across 1,804 appearances. ➤ Open weights: Drama 3 Preview is the #1 open weights model on all three of our text-to-speech leaderboards. In Pronunciation Robustness it leads the next open model, Mistral's Voxtral TTS at 78.1%, by 15 points, and in Provider Voice Arena it is ahead of Breeze TTS 2 at 1221. Drama 3 is a 35B-parameter mixture-of-experts model (3B active), supports 83 languages and is released on Hugging Face. ➤ Pricing: Drama 3 Preview is available via the Fish Audio API at $15 per 1M characters, the same as S2.1 Pro. For comparison: Eleven v4 costs $80 per 1M characters, Eleven v4 Turbo $40. Details and audio samples below 🧵
· 5.4K views
The most intelligible voice in the latest test comes from a model you can download and run yourself.
Fish Audio released the Drama 3 Preview weights on Hugging Face, and in the Artificial Analysis pronunciation test of Oct 13 the new model placed above the paid Eleven v4 and Eleven v4 Turbo. Intelligibility here means whether the voice stumbles over a name, an abbreviation or an order number.
A lead over its peers. The closest open model, Voxtral TTS from Mistral, scores 78.1% on the same test. Drama 3 Preview scores 93.2% and pulls ahead by 15 points.
Price per 1M characters. Fish Audio charges $15 for Drama 3 Preview, the same as for the earlier S2.1 Pro. Eleven v4 Turbo costs $40, Eleven v4 $80.
How lifelike the voice sounds still trails its accuracy. In the arena's blind comparisons, judges placed Drama 3 Preview 10th with an Elo rating of 1243 across 1,857 pairs. The earlier S2.1 Pro sat at 1142. In the mode with preset voice settings, the model is only 14th.
The model has 35B parameters, with 3B waking up for each request, so the giant answers at a mid-range price. It supports 83 languages, and the weights are on Hugging Face: you can run it yourself.
The release is marked as a preview; the final version has yet to be compared with Eleven v4.