All tasks / Models
Fish Audio Drama 3 (Preview)
Fish Audio's open-weights 35B mixture-of-experts text-to-speech model for multi-speaker, long-form scenes.
- Fish Audio
- Open weights
- Released 14 Oct 2026
Where it ranks
- AText to speechModels · Local · 1,244 on Artificial Analysis Speech Arena: Text to Speech (provider voices)
Facts
- Vendor
- Fish Audio
- Family
- Fish Audio Drama
- Released
- 14 Oct 2026
- Weights
- Open: download and run it yourself
- License
- Fish Audio Research License
- Parameters
- 35.9B (3B active)
- GPU memory to run locally
- ~48 GB (Official checkpoint is a 41.5 GB download (FP8 experts, other weights bf16), so a 48 GB-class GPU; an unofficial int4 build (fishaudio/drama-3-int4) is 22.1 GB. Sizes from the int4 model card.)
- API price
- $15 per 1M characters (Artificial Analysis normalised price for Fish Audio Drama 3 Preview (USD per 1M characters).)
- Access
- Free (open weights)
- Last verified
- 18 Oct 2026
Price from artificialanalysis.ai, 18 Oct 2026.
Benchmark results
| Benchmark | Result | Entry on the leaderboard | As of | Source |
|---|---|---|---|---|
| Artificial Analysis Speech Arena: Text to Speech (provider voices) | 1,244 | Fish Audio Drama 3 PreviewProvider Voice Arena, all categories and accents; rank 10 of 96, 1999 appearances, 95% CI ±15; raw value from the page's embedded data (shown rounded); no last-updated date shown, so as_of = retrieval date. | 18 Oct 2026 | artificialanalysis.ai |
Where you can use it
- AFish AudioApps · Cloud