Release 0.3 of audio.cpp introduces five new models, including Supertonic 3, MOSS-TTS, IndexTTS2, and Irodori-TTS. Supertonic 3 achieves over 200x real-time performance on an RTX 5090, enabling the generation of 10 hours of audio in approximately 3 minutes. The C++/GGML-based implementation also supports CUDA streaming with a TTFT of around 47 ms.