AssemblyAI launches tts-1: a cost-effective text-to-speech API for voice agents

AssemblyAI has launched AssemblyAI-TTS-1, its first text-to-speech model, now generally available. The API targets voice agents in live phone calls, where latency and pronunciation are critical to retaining caller engagement. Priced at $15 per 1M characters, the model offers no annual contract or concurrency fees. It closes the voice loop for developers already using AssemblyAI's speech-to-text services. TTS-1 is a streaming model trained on millions of hours of conversational audio, using continuous audio representations instead of discrete tokens, which AssemblyAI claims enhances realism and reduces latency. The launch includes 4,000+ voices, instant voice cloning, and support for six languages. Three interfaces are available, including an OpenAI-compatible route, with output in various audio formats. Performance metrics from October 28, 2026, show TTS-1 scoring 81.5% reading accuracy and 4.50 naturalness on Cekura’s benchmark. It also achieved a 159 ms median streaming latency. While Inworld TTS-2 Flash outperformed it in some metrics, AssemblyAI highlights its competitive pricing and lack of concurrency fees as key advantages.