Tasks and inputs
Drama 3 is listed for Audio generation, Text → Speech, Text-to-speech. It accepts text and produces audio. Use the publisher’s input format and preprocessing for your chosen deployment.
Fish Audio Drama 3 is an expressive 35B MoE text-to-speech model steered by bracketed directions, covering 83 languages, with downloadable weights.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Fish Audio API list price $50 per 1M characters (launch promo of $10 ended 2026-11-12). OpenRouter: $10 per 1M UTF-8 bytes. Indicative; confirm on your account.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Most text-to-speech engines read. Fish Audio Drama 3 performs. It is Fish Audio's most expressive model, released on 13 October 2026, and it takes stage directions: write a bracketed instruction and the voice follows it, either for a whole line or from mid-sentence, or for one specific word.
That changes how you write scripts. Instead of tweaking pitch sliders, you annotate the text the way a director would, and the directions do not even need to be in the same language as the spoken line.
Drama 3 is a mixture-of-experts model with 35B total parameters and about 3B active per token. It covers 83 languages and can generate multi-speaker dialogue in a single pass once speakers are assigned voices in the editor.
Fish reports 93.2% on a third-party pronunciation-robustness test, ranked first, and decoding speed of 271 frames per second on one H200. Both are vendor-reported.
The weights are downloadable from Hugging Face as fishaudio/drama-3, but under the Fish Audio Community License v1.0, not an OSI licence. In short:
Some catalogues mark it closed. It is not, but the licence terms are the real constraint.
| Item | Detail |
|---|---|
| API ID | drama-3 (one FAQ shows drama-3-preview) |
| Endpoint | /v1/tts, selected by the model header |
| List price | $50 per 1M characters |
| Launch promo | $10 per 1M characters, 13 Oct to 12 Nov 2026 (ended) |
| OpenRouter | $10 per 1M UTF-8 bytes; mp3 or pcm output |
The promotional rate ended on 12 November, so any registry figure of $10 reflects the promo, not the list price, on Fish's own API. OpenRouter's rate is per byte, not per character, so non-Latin scripts that take several bytes per character cost more there. Confirm on your account.
Some Fish pages label the model "Drama 3 Preview". Treat both names as the same model, and check which ID your account accepts.
Write a two-minute scene with two characters, an interruption, a whisper and a shouted line, and one direction in a different language from its dialogue. Render it three times. Listen for whether directions land on the right words and whether voices stay consistent across takes. Then render the same text with a plainer model, such as Gemini 3.8 Flash TTS, to see whether the expressiveness justifies the price. More models are under text-to-speech, and Fish's open releases are on the Fish Audio page.
Compare the model on the work you need to complete.
Drama 3 is listed for Audio generation, Text → Speech, Text-to-speech. It accepts text and produces audio. Use the publisher’s input format and preprocessing for your chosen deployment.
Compare pronunciation, voice consistency and time to the first playable audio. Check the service’s billing unit for the actual script.
The weights are downloadable under Fish Audio Community License v1.0. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Alibaba
Qwen-Audio-3.1-TTS Plus is Alibaba's premium text-to-speech model for brand voices and dubbing, with fine-grained emotion tags and noise-robust voice cloning.
heygen
HeyGen Voice is HeyGen's default text-to-speech model: about 39 languages, public and cloned voices, streaming with character timestamps.
heygen
HeyGen Voice Turbo is the low-latency version of HeyGen's TTS model for public voices and instant clones, at $4 per 1M characters.
Alibaba
Qwen-Audio-3.1-TTS Flash is Alibaba's streaming text-to-speech model for voice assistants, live chat and customer service, with system and cloned voices.
Microsoft
MAI-Voice-2.1 generates expressive multilingual speech from text through Azure Speech, with voice selection and supported custom-voice workflows.
Microsoft
MAI-Voice-2.1-Flash generates expressive multilingual speech from text through Azure Speech, with voice selection and supported custom-voice workflows.