TTS with a question... the audio back has the answer

Fair point. gpt-audio-1.5 is a generative audio/chat-style model, so if the goal is strict verbatim narration, it can still require careful prompting and delimiter handling. For this specific /v1/audio/speech issue, the most reliable path is to stay on Speech API models. If literal readout matters more than expressive voice control, try tts-1 or tts-1-hd. If you need the newer controllable TTS behavior, use gpt-4o-mini-tts, but the 2025-12-15 snapshot is the one showing this prompt-like interpretation issue. So the practical split is: tts-1 / tts-1-hd for stricter read-the-text behavior, gpt-4o-mini-tts for more expressive speech control. Hope this helps.