TTS leaderboard: xAI hits 87.6% pronunciation accuracy, Kokoro 82M fastest at 242 chars/s

ArtificialAnlys · x · 2026-09-22

Artificial Analysis launched a full Text-to-Speech model comparison covering Arena Elo, speed, price, and a new Pronunciation Robustness metric (human-judged correctness of English spans), ranking 27 of 93 models. Key numbers: xAI TTS scores 87.6% accuracy at 106 chars/s; Realtime TTS-2 Flash hits 77.8% at 220 chars/s; open-source Kokoro 82M v1.0 is fastest at 242 chars/s but scores just 55.5%.

Related event: Artificial Analysis Launches TTS Pronunciation Robustness Benchmark(6 posts)→

Original post →

More from Models

Models channel →