TTS-1

TTS-1 arrived on 6 November 2023 as one half of OpenAI's first text-to-speech pair, alongside the higher-quality TTS-1 HD. It converts written text into speech through the Audio API's speech endpoint and is tuned for real-time use, where latency matters more than the last few percent of audio fidelity. Six preset voices ship with it, and OpenAI states plainly that the voices are artificial presets it monitors, not clones of real people. The model is deliberately narrow: text in, audio out, no reasoning, no context window, no tool use. That narrowness is also why it has aged well enough to stay on the price list. OpenAI now sells newer speech models built on its multimodal line — gpt-4o-mini-tts and the realtime family — which can be instructed on how to say something, not just what to say. TTS-1 cannot. What it offers instead is a fixed, predictable cost per character and a well-understood latency profile. Its pricing is worth stating precisely because it is unusual. At launch OpenAI quoted $0.015 per 1,000 input characters; the current price list quotes $15 per million characters. Those are the same number. Almost three years of model releases, price cuts across the text line and two generations of audio models have gone by without this model getting cheaper or more expensive.

▸Videos

No videos yet.

▸Reviews