---
title: "How Much Does Fish Audio Drama 3 Cost? Pricing and License"
url: "https://goremy.ai/blog/articles/fish-audio-drama-3-pricing-license-api"
author: Luis Chavez-Mattos
editor: Luis Chavez-Mattos
editor_title: Director, Product
published: "2026-10-23T15:00:10.365Z"
updated: "2026-10-23T15:00:10.365Z"
topic: AI Pricing & Usage Limits
description: "Fish Audio Drama 3 API costs about $10 per million characters at launch, then $50 after Nov 12. License limits, fallback billing and free trial explained."
---

# How Much Does Fish Audio Drama 3 Cost? Pricing and License

> Fish Audio Drama 3 API costs about $10 per million characters at launch, then $50 after Nov 12. License limits, fallback billing and free trial explained.

Edited by Luis Chavez-Mattos, Director, Product · Published 2026-10-23

## How much does Fish Audio Drama 3 cost?

Through the API, Drama 3 costs about $10 per million characters right now. That is a launch discount, and the price jumps to $50 per million characters after November 12th. The model weights are also open on Hugging Face, and the license is free for commercial use if you make under roughly $500K a year. You can try the model for free on Fish Audio's Drama 3 page before paying anything.

## TL;DR

- The API runs about **$10 per million characters at launch**, then rises to $50 after November 12th, a fivefold increase.
- The license is **free for commercial use under about $500K a year**, which covers most solo creators and small studios.
- The API can **quietly fall back to older models** and still bill you, so set the model to Drama 3 explicitly and check your usage.
- **Takes vary** from run to run, so budget for several generations per line and keep the best one.
- Self-hosting is possible but heavy: the model has **36 billion parameters** and reportedly needs around a 46 GB VRAM card.
- A **free web trial** lets you test direction tags and multi-speaker scripts before you commit to anything.

## What is Fish Audio Drama 3?

Drama 3 is a text-to-speech model from Fish Audio. Instead of picking an emotion from a dropdown (happy, sad, angry), you write performance direction in plain English inside square brackets in the script. A line might carry a note like "barely holding it together, forcing a smile." You can aim direction at specific words, such as telling the model to whisper one phrase. You can also place sound effects like a small laugh or a shaky breath exactly where you want them, and script multiple speakers with different voices.

On third-party measurement, it looks strong in one area and mixed in another. On Artificial Analysis, Drama 3 scores highest on pronunciation, ahead of ElevenLabs and SpaceX AI's text-to-speech model. On the arena leaderboard, where listeners pick favorites in blind comparisons, it did not do as well. It reads hard words better than almost anything, but many listeners still prefer other voices.

## What does the API pricing look like over time?

The numbers come from a creator's summary of the launch, so confirm them on Fish Audio's pricing page before you plan around them.

| Period | Approx. price per 1M characters |
|---|---|
| Launch (now) | $10 |
| After November 12th | $50 |

To make that concrete, here is the arithmetic. A script of 500,000 characters would cost about $5 at the launch rate and about $25 after the increase. That is small money for a single project. It matters more if you generate at volume, such as a long audiobook catalog, a game with thousands of dialogue lines, or a product that voices user content.

If you have a batch job planned, the deadline is the practical takeaway. Anything you can generate before the price change costs a fifth of what it will afterward.

## Why do take variability and fallback billing matter?

Two quirks can inflate a bill without much warning.

**Takes vary.** Fish Audio's output is not deterministic in a way you can rely on. The same line with the same direction can come out noticeably different on each run. The practical advice is to generate a few versions of important lines and keep the best. That multiplies your character count. If you render each line three times, your real cost is three times the per-character rate applied to your script length.

**The API can fall back to older models.** If you call the API and do not set the model to Drama 3, requests can quietly route to an older Fish Audio model. You still get billed, and you may not notice, because the audio still comes back. The result is that you can pay for output that lacks the direction-following behavior you wanted.

A few habits help:

1. Set the model parameter to Drama 3 explicitly in every request.
2. Run a short test with a bracketed direction and confirm the output actually follows it.
3. Check your usage dashboard after the first batch, not after the whole project.
4. Log which model each request used if your pipeline allows it.

## What are the license limits?

The reported terms are simple. The license is free for commercial use as long as you make under about half a million dollars a year. For most independent creators, indie game developers and small audiobook producers, that threshold is unlikely to bind.

If your business is larger, read the license text itself. The summary available here does not spell out what happens above the threshold, whether the cap applies to revenue or some other measure, or how it treats products built on top of the model. Do not assume terms beyond what the license says.

One related note: Fish Audio licensed Uncle Roger's voice for its demo page. That is a licensed celebrity voice used in a showcase, not something the general license hands you. Cloning or imitating real people raises its own rights questions that are separate from the model license.

## Can you run Drama 3 yourself instead of paying for the API?

Probably, if you have the hardware. The weights appear to be open on Hugging Face. The model has 36 billion parameters, so it is heavy. People have reported needing roughly a 46 GB VRAM card to run it as released. A community 4-bit quantized version reportedly fits in about 32 GB. Neither will run on a typical laptop.

Self-hosting changes the cost model. You trade per-character fees for GPU rental or purchase, and the launch discount stops mattering. The math favors self-hosting when you have steady, high-volume generation and a GPU you already own or can rent cheaply. For occasional use, the API is simpler. Remember that self-hosting does not remove the license terms, and it does not remove take variability. You still generate multiple versions to find the good one, though now the extra takes cost compute time instead of API credits.

## Is Drama 3 worth trying?

If you produce anything with characters, probably yes. Audiobooks, game dialogue and scripted multi-speaker audio benefit most from per-line direction and inline sound effects. Tag-based control is more flexible than an emotion dropdown, and the pronunciation scores suggest it handles difficult words well.

The tradeoffs are real. Blind listening tests did not rank it first, so your own ears should decide. Costs rise sharply after November 12th. And you will spend extra generations on retakes. The free trial lets you test your own scripts, including bracketed direction and multi-speaker scenes, before you build a pipeline around it.

## Frequently Asked Questions

### How much does the Drama 3 API cost per million characters?

About $10 per million characters at the current launch discount. After November 12th it rises to $50. Verify the figures on Fish Audio's pricing page, since promotional rates can change.

### Can I use Drama 3 commercially for free?

According to the reported terms, yes, if you make under about half a million dollars a year. Above that, you should read the license text or contact Fish Audio, because the available summary does not describe the terms for larger businesses.

### Why was I billed when I did not get Drama 3 output?

The API can quietly fall back to older Fish Audio models if the model is not set to Drama 3, and those requests still get billed. Set the model explicitly in each request and check your usage after a small test batch.

### Can I try Drama 3 without paying?

Yes. Fish Audio's Drama 3 page has a free trial option, so you can test direction tags, whispered words, sound effects and multi-speaker scripts before touching the API.

### What hardware do I need to run Drama 3 locally?

Reports put the full model at roughly 46 GB of VRAM. A community 4-bit quantized build reportedly fits in about 32 GB. The model has 36 billion parameters, so most laptops are ruled out.