Fish Audio models, access and benchmark guide
Fish Audio model and access guide with a practical US-English evaluation brief and links to publisher materials.

👁 111 reads
Drama and live conversation are different deliverables in Fish Audio’s lineup. A dramatic narration depends on pacing and character separation, whereas a live widget depends on interruption and the cost of an open session. Each deserves its own listening fixture.
Choosing the first job
Record an original dialogue with alternating lines, then interrupt a separate dummy agent call and check recovery.
Models in the November 17 snapshot
| Model and track | Placement | Reported score |
|---|---|---|
| Fish Audio Drama 3 (AA provider voice) | 7 | 1279 Elo |
| Fish Audio Drama 3 (AA controlled voice) | 17 | 1070 Elo |
| Fish Audio Drama 3 Preview (AA provider voice) | 13 | 1236 Elo |
| Fish Audio Drama 3 Preview (AA controlled voice) | 23 | 1059 Elo |
| Fish Audio S2.1 Pro (Oct 2026) (AA provider voice) | 25 | 1181 Elo |
| Fish Audio S2.1 Pro (Oct 2026) (AA controlled voice) | 33 | 1008 Elo |
Release record. Publisher documentation identifies the product family. The ranked configurations include Fish Audio Drama 3, Fish Audio Drama 3 Preview, Fish Audio S2.1 Pro (Oct 2026).
Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →Sources
- https://docs.fish.audio/ (accessed November 17, 2026)
- https://github.com/fishaudio (accessed November 17, 2026)
- https://artificialanalysis.ai/text-to-speech/leaderboard/provider-voice (accessed November 17, 2026)
- https://artificialanalysis.ai/text-to-speech/leaderboard/controlled-voice (accessed November 17, 2026)
More on this from QUANTSLANT
benchmarks·fal models, access and benchmark guidefal model and access guide with a practical US-English evaluation brief and links to publisher materials.