ModelsNews
AssemblyAI-TTS-1 launches at $15 per million characters
AssemblyAI, known for its speech-to-text models, has released its first text-to-speech model. AssemblyAI-TTS-1 targets voice agents: $15 per million characters, unlimited concurrency and no contract.
By: Mikola Máté Patrik · · 2 min read

What is AssemblyAI-TTS-1?
AssemblyAI introduced its first text-to-speech model on October 28 (AssemblyAI). The company has been strong in speech-to-text and now covers the output end of the voice agent pipeline too. The model uses continuous audio representations instead of discrete audio tokens, which the company says produces more natural speech.
- Pricing: $15 per million characters, unlimited concurrency, no contract.
- More than 4,000 prebuilt voices, instant voice cloning and six languages: English, Spanish, French, German, Italian and Portuguese.
- HTTP, WebSocket streaming and an OpenAI-compatible API, in US or EU regions.
What do the benchmarks show?
By the company's own measurements, median time to first byte is 47 ms and first audio arrives in about 90 ms in production. Entity accuracy (names, numbers, addresses) is 93.7% versus 93.6% for ElevenLabs v4, word accuracy is 99.0% and MOS is 3.85.
MarkTechPost also cites an independent test: Cekura measured 159 ms streaming latency and 81.5% reading accuracy (MarkTechPost). Self-reported and independent numbers differ, so test it on your own text.
What is missing
At launch there is no SSML, speed control, pronunciation dictionaries, emotion tags or Pro-level voice cloning. Inworld TTS-2 Flash costs the same, so competition will come down to latency and quality.
What it means for you
If you build voice agents and already use AssemblyAI for STT, you can keep the whole speech pipeline with one vendor. The OpenAI-compatible API makes a trial a small code change.
Produced with the help of AI, reviewed by an editor. How we make our articles
Reviewed and edited by the Neural Daily desk. About


