Fish Audio S2.1 Pro
Pay-as-you-go voice cloning — 70% cheaper than ElevenLabs
Comparative Scores
Architecture
Priced production Fish model. For Où est Ava ? emotional direction, use the Drama 3 card: s2.1-pro stays the model with a published rate and the production recommendation.
Analysis
Fish Audio’s priced production model is S2.1 Pro: $15 per million UTF-8 bytes, 83 languages on the models overview, free-form bracket directions and a live WebSocket. OpenAudio S1 is deprecated and retires on 31 December 2026, after which it is served as S2.1 Pro. The public arena on 17 October 2026 lists S2.1 Pro at ELO 1190, rank 18. Drama 3 is a separate preview card and has no line on the rate card.
Strengths
- S2.1 Pro ELO 1190, rank 18 (public arena, 17 Oct 2026)
- Voice cloning at the same API price
- $15 / 1M UTF-8 bytes, pay as you go
- 83 languages on the models overview
- s2-pro is documented as open-source; Drama 3 is a separate preview
Weaknesses
- ~200ms TTFA (slower than Cartesia)
- No native lip-sync data
- Limited documentation
Voice Capabilities
Voice cloning included at no extra cost. 10-second audio sample. 70% cheaper than ElevenLabs for equivalent quality.
S2.1 Pro interprets free-form bracket directions. Tags that change only a few words live on Drama 3.
How: Natural-language `[bracket]` directions, sounds and pauses. The emotions doc still lists a named repertoire (`[happy]`, `[whispering]`, `[break]`).
Validate: This is not the Drama 3 contract. For an Où est Ava ? line that must whisper only one clause, test the Drama 3 card.
Streaming API available. ~200ms TTFA.
No native lip-sync timestamps.
Pricing
$15/1M chars. Pay-as-you-go, no subscription. Voice cloning included at same price.
Sovereignty & Compliance
Cloud API. S1-mini weights available for self-hosting.
Data residency: US/Asia
Fish Audio S2.1 Pro — Strategic Positioning
Beyond technical specs: where does this tool sit in the ecosystem, what are the risks and strategic implications for GamiWays?
Fish Audio is the open-source disruptor: enterprise-grade voice cloning with natural language emotion control, available both as a cloud API and a self-hostable model — the clearest path from Phase 1 speed to Phase 2 sovereignty.
A. Strategic Positioning
Target customer: Developer / SMB / Enterprise — voice cloning, multilingual content
Open-source S2 model with cloud API — enterprise-grade voice cloning and expressiveness at a fraction of ElevenLabs cost.
B. Competitive Moat
- Open-source S2 model (self-hostable) + cloud API — dual deployment flexibility
- Natural language emotion tags — superior expressiveness control vs SSML
- 50% enterprise revenue share — strong institutional adoption signal
Vulnerability: No explicit compliance certifications (SOC2, HIPAA). Fast-moving open-source market could be disrupted by better-funded competitors.
E. Strategic Questions for GamiWays
Sovereignty fit
Open-source S2 model enables full self-hosting in Swiss/EU infrastructure. No compliance certs for cloud API, but self-hosted path is clear.
Build vs. Buy
Build (self-host S2) for Phase 2 sovereignty. Buy (cloud API) for Phase 1 speed. Best of both worlds.
Lock-in risk
Open-source S2 model eliminates vendor lock-in. Cloud API has moderate lock-in but self-hosted alternative always available.
Roadmap alignment
Excellent: cloud API for Phase 1 speed, self-hosted S2 for Phase 2 sovereignty. Natural migration path.
This sheet's verification
Artificial Analysis Speech Leaderboard + Fish Audio docs, 2026
Update note: Aperçu des modèles du 18 octobre 2026 : s2.1-pro, 83 langues, 15 $ / million d’octets UTF-8. s1 retiré le 31 décembre 2026. Arena du 17 octobre : ELO 1190, rang 18. Drama 3 est une fiche séparée, absente de la grille.
This date covers provider-specific prices, capabilities and notes. Comparative benchmarks follow the synchronization and methodology shown above.
ELO benchmarks / indices: Artificial Analysis · Last API sync : 5 October 2026