Fish Audio Open-Sources Drama 3: Built on Qwen3.5-35B-A3B, Focused on Fine-Grained Voice Control

Beating AI News Flash: AI voice company Fish Audio has officially open-sourced the Drama 3 model weights, allowing developers to download from Hugging Face and deploy on their own. Drama 3 launched a preview version as early as September, and this time it is fully opening its web, App, and API, while also publicly releasing the model weights, inference code, and technical report.

Drama 3 is built on the Qwen3.5-35B-A3B language model, with a total parameter count of 35 billion, activating only about 3 billion parameters per generation, and supporting 83 languages. It focuses on fine-grained control of voice performance, allowing users to directly describe emotions in natural language, making a line shift from calm to choked with emotion, inserting laughter at specified positions, and even letting up to 16 characters take turns speaking within the same passage of text.

In Artificial Analysis's English pronunciation accuracy test, Drama 3 ranked first with 93.2%, higher than the previous generation S2.1 Pro's 88.1%.

Drama 3 uses the Fish Audio Community License. It is free for research and non-commercial use. Individuals or enterprises and their affiliates with total annual revenue below $500,000 can use it commercially for free within the permitted scope. However, if the model is provided as a voice generation, cloning, or inference API service to others, separate authorization is still required.