OpenAI Text-to-Speech

OpenAI's speech endpoint turns text into audio with thirteen named voices and, in its newest model, a free-text instruction for tone and accent. Developers who already build on OpenAI's chat models use it for narration or an assistant voice without a second vendor. Arabic is supported only in the sense that the documentation says language coverage "generally follows" the Whisper transcription model, which includes Arabic; the voices themselves are described as optimized for English, there is no Arabic-specific voice, no dialect option and no voice cloning. On 1 October 2026 OpenAI announced that its current text-to-speech models will be retired on 6 January 2027, with developers directed to the Realtime API's speech output instead, so anything built on it today has a known end date. Pricing in November 2026: the retiring models no longer appear on the public pricing page; the successor is billed per audio output token at $20 per million through the Realtime API.