tts-1

Overview. tts-1 is OpenAI's dedicated text-to-speech model designed for real-time speech synthesis. Released in November 2023, it converts text input directly into spoken audio across nine built-in voices. The model is served as a proprietary API and is scheduled for retirement in January 2027 in favor of OpenAI's newer real-time speech architectures.

Key strengths. The model is optimized for minimal synthesis latency, enabling rapid audio turnaround for interactive voice applications. It natively exports to multiple standard audio formats, including MP3, Opus, AAC, FLAC, WAV, and raw PCM.

When to use it. Use tts-1 for real-time voice agents, screen readers, and interactive dialogue systems where latency is more critical than studio-grade acoustic fidelity. It fits direct text-to-speech pipelines handling text blocks of up to 4,096 characters per request.

Limitations. Individual synthesis requests are strictly limited to 4,096 characters, requiring client-side chunking for long-form text. Audio fidelity is intentionally compromised for generation speed compared to dedicated high-definition alternatives. The model has been formally deprecated by OpenAI, with an end-of-life date of January 6, 2027.

Sovereignty note. tts-1 is hosted by US-based OpenAI, subjecting inference to US regulatory jurisdiction and CLOUD Act exposure. It is not hosted on Sovereign European infrastructure.