Fish Audio's open-weights Drama 3 Preview tops TTS leaderboards, beats Eleven v4 on pronunciation
ArtificialAnlys · x · 2026-10-14
Artificial Analysis reports that Fish Audio's open-weights TTS model Drama 3 Preview is now the top open-weights model on its Text to Speech leaderboard and ranks #1 overall on Pronunciation Robustness at 93.2%, ahead of Eleven v4 (91.7%) and Eleven v4 Turbo (90.1%).
- Built for multi-character scenes and long-form content with voice cloning and natural-language control; a 35B-parameter MoE (3B active) supporting 83 languages, weights on Hugging Face under the Fish Audio Community License
- Ranks #1 in Standalone Terms (95.9%) and Preserving Exact Sequences (86.3%), #2 in Expanding Shorthand (93.6% vs Eleven v4's 94.1%)
- Provider Voice Arena Elo of 1,243 (#10), a 101-point jump over the previous Fish Audio S2.1 Pro; leads the next open-weights model, Mistral's Voxtral TTS, by 15 points on pronunciation
- API pricing at $15/1M characters, same as S2.1 Pro, versus $80 for Eleven v4 and $40 for Eleven v4 Turbo
Related event: Fish Audio Open-Sources Drama 3, Topping TTS Leaderboards(11 posts)→
More from Multimodal
- Fish Audio open-weights Drama 3 TTS with word-level tone control after $52M seed — azed_ai · 2026-10-14
- World Labs-generated scenes run smoothly in AR via WebGPU: a parallax portal cube demo — yungcontent · 2026-10-14
- Noob Tip: Extract the Exact ComfyUI Workflow Back Out of Your Generated Images — zMarceeh · 2026-10-14
- Gemini Argon 4 High's Pelican Riding a Bicycle SVG — Thewolf421 · 2026-10-14
- FastH3-Omni lands Oct 20 with open weights, local RTX/Spark/MLX support — Vandy_simp · 2026-10-14
- AI short 'The Last Span' wins Astana fest award; video gen took just 1 of 6 weeks — lmoroney · 2026-10-14