Fish Audio Drama 3 is an open-weight 35B text-to-speech model for studios and creators who need expressive, multi-character voice work they can direct like an actor. Released on October 13, 2026, it speaks 83 languages and lets you write stage directions in plain brackets that change the delivery from that point onward.
Core Features
- 35B-parameter MoE backbone with about 3B active per token, initialized from Qwen3.5-35B-A3B.
- Bracketed direction such as [choked up, crying quietly] applied mid-line, plus paired tags like emphasis or whisper that scope only selected words.
- Up to 16 speaker tags in one script, so a full dialogue scene renders as a single stream.
- Automatic long-text chunking and streaming for audiobooks and dubbing.
- Ranked #1 on Artificial Analysis' Pronunciation Robustness benchmark at 93.2 percent, ahead of Fish's own S2.1 Pro at 88.1 percent.
Use Cases and Best For
- Audiobook and audio-drama producers who need consistent character voices across long scenes.
- Localization teams dubbing content into many languages with fine mood control.
- Creators wanting free commercial voice generation below the revenue cap.
Pricing
Weights are free under the Fish Audio Community License v1.0 for commercial use below US$500,000 annual revenue; hosted voice services need a separate licence. The API lists at $50 per million characters, discounted to $10 through November 12, 2026.
Our Take
Best for expressive narration where direction matters more than raw speed; the trade-off is a roughly 40 GB download and invite-only in-place editing.
Compare more voice models in our AI audio tools directory.
