Fish Audio Drama 3 Pricing: API Costs, Free Tier, and the Nov 12 Deadline
Fish Audio Drama 3 pricing explained: $10 per million characters before November 12, free commercial use under $500K revenue, and trial access.
What does Fish Audio Drama 3 cost right now?
Fish Audio’s Drama 3 text-to-speech model costs about $10 per million characters through the API, but that’s a launch discount. The price jumps to $50 per million characters after November 12. The model is also free to try directly on Fish Audio’s site, and it carries a free commercial-use license for companies making under $500,000 a year in revenue. If you’re building anything with voice, audiobooks, game dialogue, narration, that pricing window matters more than the feature list.
TL;DR
- Launch pricing for Drama 3 sits at roughly $10 per million characters through the API, a five times discount off the post-November 12 rate of $50 per million characters.
- The commercial license is free for companies or individuals generating under $500,000 in annual revenue, which covers most solo builders and small studios.
- A free trial is available directly on Fish Audio’s site at the Drama 3 page, no API key needed to test the voice directing features.
- The model defaults can silently fall back to older Fish Audio models if you don’t explicitly set the model parameter to Drama 3, and you’ll still get billed for those generations.
- Drama 3 is open weight on Hugging Face, a 36 billion parameter model that needs serious hardware (roughly 46GB VRAM, or 32GB with a community 4-bit quantized version) to self-host.
- Pronunciation benchmarks are strong, with Drama 3 scoring highest among tested models on Artificial Analysis for reading difficult words, even ahead of ElevenLabs.
- Human preference scores are more mixed, meaning accurate pronunciation didn’t translate into the top spot on blind listening tests.
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
Why does the pricing change on November 12?
Fish Audio is running a standard launch discount strategy: price the API low to get developers testing and integrating Drama 3, then raise it once adoption builds. The jump from $10 to $50 per million characters is a 5x increase, which is steep enough that anyone planning to use this at volume should lock in usage or at least budget around that date. If you’re prototyping a product that leans on text-to-speech, generating test content now, before the cutoff, costs a fraction of what it will cost in a few weeks.
This kind of pricing cliff is common with newly launched AI APIs, but a 5x jump is on the aggressive end. Worth checking Fish Audio’s pricing page directly before building anything that assumes the discount rate is permanent.
How much does the free commercial license actually cover?
Fish Audio allows free commercial use of Drama 3 as long as the user or company stays under $500,000 in annual revenue. For most indie developers, solo creators, and small studios, that threshold is a non-issue. It means you can ship a product, an audiobook, a game, a narrated video series, using Drama 3 output without paying for a commercial license, as long as your business stays under that revenue line. Once you cross $500K annually, you’d need to look at Fish Audio’s paid licensing terms, though specifics of that tier weren’t detailed.
This structure is similar to how several open-weight model providers handle commercial licensing: free below a revenue threshold, negotiated or paid above it. It lowers the barrier for experimentation and small-scale shipping while still capturing revenue from larger commercial users.
What makes Drama 3 different from other TTS tools?
Most text-to-speech tools hand you a dropdown menu of preset emotions, happy, sad, angry, and you pick one. Drama 3 instead lets you write direction in plain English, placed in square brackets directly inside the script, the way you’d note stage directions for an actor. A line like “I’m fine, really. Go on without me” can be tagged with direction like “barely holding it together, forcing a smile,” and the model adjusts delivery accordingly.
It goes further than tone. You can target specific words for emphasis or whispering, drop in sound effects like a small laugh or a shaky breath at an exact point in the line, and generate multi-speaker dialogue with distinct voices for each character. Fish Audio also licensed a recognizable internet personality’s voice (Uncle Roger) for its demo page, which gives a sense of how convincingly the model can replicate a specific, textured speaking style.
Is Drama 3 actually good, or just good at pronunciation?
On Artificial Analysis benchmarks, Drama 3 scored highest on pronunciation accuracy among tested text-to-speech models, including ahead of ElevenLabs and other established players. That’s a real, measurable win for reading difficult words correctly.
One coffee. One working app.
You bring the idea. Remy manages the project.
But pronunciation accuracy and listener preference aren’t the same thing. On the Arena leaderboard, which runs blind preference tests where people listen to outputs from different models without knowing which is which, Drama 3 didn’t come out on top. People still preferred other voices in direct comparisons, even when Drama 3 was technically more accurate. The takeaway: good pronunciation doesn’t automatically mean a more natural or appealing voice. If you’re picking a TTS model for a product, it’s worth generating a few outputs from Drama 3 and comparing against alternatives for your specific use case, since quality can vary take to take with this model.
Can you run Drama 3 yourself instead of using the API?
Yes. Drama 3 is open weight and available on Hugging Face, but it’s a 36 billion parameter model, which means you need real hardware to run it locally. Reports suggest you need around 46GB of VRAM to run the full model, or you can get by with roughly 32GB using a community-made 4-bit quantized version. Neither of those specs will run on a typical laptop or consumer desktop GPU without a high-end card. For most developers, the API is the more practical entry point unless you already have access to a workstation or cloud GPU with that much memory.
What should you watch out for when using the API?
The most practical gotcha reported with Drama 3’s API: it can quietly fall back to older Fish Audio models if you don’t explicitly specify Drama 3 as the model in your request. You’ll still get billed, but you won’t get the Drama 3 features or quality you were expecting. Double-check your model parameter on every call, especially if you’re integrating it into a pipeline where this setting might get overlooked or defaulted.
It’s also worth generating multiple takes of any critical line. Output quality can vary between generations, so plan to generate a few versions and keep the best one rather than treating the first output as final.
Frequently Asked Questions
How much does Fish Audio Drama 3 cost per million characters?
Around $10 per million characters at launch pricing, rising to $50 per million characters after November 12.
Is Fish Audio Drama 3 free to use?
You can try it for free directly on Fish Audio’s site without an API key. Commercial use is also free if your revenue is under $500,000 per year.
Can I run Drama 3 on my own hardware?
Yes, it’s open weight on Hugging Face as a 36 billion parameter model. You’ll need roughly 46GB of VRAM, or about 32GB with a community 4-bit quantized version, which rules out most laptops and consumer setups.
Does Drama 3 sound better than ElevenLabs?
It scores higher on pronunciation accuracy in Artificial Analysis benchmarks, but blind preference testing on the Arena leaderboard shows mixed results, with listeners not consistently favoring it over other models.
Why would my API bill show charges for a different model?
Fish Audio’s API can default to an older model if you don’t explicitly set the model parameter to Drama 3. Always verify that setting before running production jobs to avoid unexpected fallback billing.