S2.1 Pro

AlphaSignal reported Fish Audio's S2.1 Pro free API launch, adding technical detail beyond the official announcement: an Audio Turing Test score of 0.515 (surpassing Seed-TTS by 24% and MiniMax-Speech by 33%), and a description of Fish Audio's custom "fish-scales-ops" FP8 GPU kernels reported as 2.1–4.3x faster than th Migration friction is described as minimal: existing Fish Audio users switch by setting model "s2.1-pro-free" in their API headers using the same endpoint. AlphaSignal states the free access window runs through the end of July 2026, with businesses over $1M ARR required to contact Fish Audio before using the free tier

Fish Audio announced that its S2.1 Pro text-to-speech model is now available as a free API with unlimited usage under its Fair Use Policy, with a model string of "s2.1-pro-free" that drops into existing Fish Audio calls. According to the Fish Audio blog, S2.1 Pro is the company's state-of-the-art voice model, building Performance details from the Fish Audio blog include a 61% win rate against the prior-generation S2 Pro in head-to-head listening evaluations, a Time-to-First-Audio (TTFA) of 70ms on single requests (down from 100ms), and 2x+ throughput improvement under high-concurrency load. The model supports 83 languages. The creat

The Coval benchmark dashboard independently measures S2.1 Pro across TTS models on 30-day averages, explicitly naming the variant. S2.1 Pro ranks 18th out of 30 models on Time-to-First-Audio at 337 ms and approximately 10th on Word Error Rate at 4.8%. A separate "S2.1 Pro Free" variant is measured at 1066 ms TTFA (rank The measured 337 ms TTFA on shared infrastructure is substantially higher than Fish Audio's claimed 70 ms single-request TTFA, likely reflecting shared-endpoint routing versus single-request conditions. Sample sizes are partial in the excerpt but visible (e.g., thousands of samples per model). Creator is Fish Audio; th