Inworld AI announced that Realtime TTS-2 reached general availability. The company states the update follows heavy usage of the research preview and marks its biggest model leap to date. Artificial Analysis confirmed the model now ranks first on its Controlled Voice Arena, ahead of Cartesia Sonic 3.6. Multiple observers noted support for plain-language delivery instructions such as emotion tags or cues like [sigh] and [laugh], plus multi-turn context. Inworld's own page describes the system accepting full audio input and maintaining one voice identity across more than 200 languages while targeting sub-100 ms latency.