OpenAI TTS-1

openai/tts-1 · by OpenAI · neural-tts

Pricing — 1 offering(s)

OpenAI (Audio)

About this offering: OpenAI TTS-1 (text-to-speech). Standard quality, optimised for speed and latency. Available via the /v1/audio/speech endpoint. Supports 6 preset voices. No voice cloning or fine-tuning. Priced at $15/1M characters with no subscription required — pure pay-per-use. Deprecated: OpenAI removes TTS-1 from the API on January 6, 2027 and recommends gpt-realtime-2.1-mini instead, so plan new work on a different model.

Text characters

$15.00 / 1M characters Current 2023-11-06 → 2027-01-05

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

See how OpenAI TTS-1 fits into a cost-aware routing setup

About OpenAI TTS-1

OpenAI TTS-1 is a proprietary neural text-to-speech model explicitly optimised for latency over fidelity. It offers 6 preset voices across 57 languages with chunked streaming through the API, and deliberately omits voice cloning, custom voice creation, and SSML — prosody control is limited to what the plain API exposes.

Its role in OpenAI's lineup is the low-cost option: at $15 per 1 million characters it is the cheapest major TTS API, with TTS-1 HD as the quality-tier sibling for voiceover and narration work where naturalness matters more than price or speed.

On independent measurement (checked 2026-08-19), the Artificial Analysis Text-to-Speech Arena ranks "TTS-1" #34 of 98 voices tracked (Elo 1094) — mid-table, just behind TTS-1 HD (#30, Elo 1108) — which is consistent with Modelglass's "moderate" naturalness rating. There is no independently published MOS for it, and OpenAI does not disclose specific first-byte latency figures.

Capability profile

How OpenAI TTS-1 rates across core capability dimensions, with the task-level evidence behind each rating.

voice naturalness Moderate

Clear, natural-sounding speech for most content, with slight artifacts on complex phonemes. In the Artificial Analysis Text to Speech Arena (checked 2026-08-19), TTS-1 ranks #34 of 98 voices (Elo 1094), just behind TTS-1 HD (#30, Elo 1108).

language support Strong

Supports 57 languages. Quality varies by language; English is strongest.

voice variety Weak

Only 6 preset voices; no voice cloning or custom voice creation.

streaming latency Strong

Low latency; chunked streaming supported via the API.

cloning support Weak

No voice cloning, no fine-tuning, no SSML support. Preset voices only.

Benchmarks

Benchmark	Score	Config	Source
MOS (naturalness)	—	No independently published MOS. Vendor positions as latency-optimised; community rates quality as moderate.	—
First-byte latency	—	Vendor-described as low latency. Specific ms values not publicly disclosed by OpenAI.	—

Operator guidance

Deprecated: OpenAI removes TTS-1 from the API on January 6, 2027 (use gpt-realtime-2.1-mini), so don't build new long-term work on it. Until then, choose TTS-1 when cost and speed are the priority. At $15/1M chars it is the cheapest major TTS API. Upgrade to TTS-1 HD for quality-critical use (voice- overs, narration). Use ElevenLabs for natural voice quality or voice cloning.