Skip to content
Inworld AI· inworld·· 2026-10-22AI score33

Cekura releases TTS benchmark, Inworld's TTS-2 series leads in latency and accuracy

@cekuraAi's new Text-to-Speech Bench tested 26 models across 30,564 requests.

AI summary

Cekura's text-to-speech benchmark covers 26 models and 30,564 requests. Inworld's TTS-2 Flash has a median latency of 46ms from receiving a complete sentence to starting speech, and TTS-2's estimated read accuracy is 93.4%. The test used 104 sentences, with each model reading them three times in the same cloud region, and accuracy was estimated by an automatic speech-to-text scorer. TTS-2 Flash's estimated accuracy is 93.2%, and the test content included phone numbers, dates, code, and medical terms.

Source: Inworld AI · x.com