Text-to-speech model powered by GPT-4o mini.
GPT-4o Mini TTS is OpenAI's cost-efficient text-to-speech model, released on March 20, 2025 as part of a next-generation audio model suite alongside gpt-4o-transcribe and gpt-4o-mini-transcribe. Built on the GPT-4o multimodal foundation, it converts text input into natural-sounding speech at approximately $0.015 per minute of generated audio — making it one of the most affordable quality TTS options on the market. The model supports 11 built-in voices (Alloy, Ash, Ballad, Coral, Echo, Fable, Nova, Onyx, Sage, Shimmer, Verse) covering a wide range of tones from deep authoritative narration to bright energetic marketing, with additional voices Marin and Cedar added through a later update.
The defining feature of GPT-4o Mini TTS is its steerability: a unique instructions parameter lets developers control not just what the model says but how it says it — using natural language prompts to specify emotion (excited, calm, urgent), pace (faster, slower), accent, or character direction. This is a significant differentiator from traditional TTS systems (tts-1, tts-1-hd, ElevenLabs) where voice characteristics are more fixed. The model supports 50+ languages, synchronous and streaming modes, and integrates seamlessly with the OpenAI SDK as a drop-in replacement at the /v1/audio/speech endpoint — fully accessible through TheRouter's unified API at api.therouter.ai/v1.
instructions parameter.| Type | Rate |
|---|---|
| Text | $16.20 / 1M chars |
| Input | $0.648 / 1M tokens |
| Output | $12.96 / 1M tokens |
Billing unit differs from the vendor's: OpenAI prices gpt-4o-mini-tts per token (text input $0.60/MTok, audio output $12.00/MTok), while this model is charged per 1M characters — the unit it is settled in here. Only tts-1 and tts-1-hd are genuinely per-character at the vendor.
| Release date | March 20, 2025openai.com ↗ | verified |
| Base architecture | GPT-4o mini multimodal foundation, fine-tuned for audio generationopenai.com ↗ | verified |
| Endpoint | /v1/audio/speechdevelopers.openai.com ↗ | verified |
| Voices | 11 core (Alloy, Ash, Ballad, Coral, Echo, Fable, Nova, Onyx, Sage, Shimmer, Verse) + later additions (Marin, Cedar)docs.aimlapi.com ↗ | verified |
| Supported languages | 50+tokenmix.ai ↗ | verified |
| Context window | ~2,000 input tokenstokenmix.ai ↗ | to verify |
| Output formats | MP3, Opus, AAC, FLAC, WAV, PCMdevelopers.openai.com ↗ | verified |
| Steerability | Yes — instructions parameter for emotion, pace, accent, and character direction via natural language promptssimonwillison.net ↗ | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
Steerability benchmark (Simon Willison) | Confirmed — instructions parameter works as designed; occasional instruction-leak edge cases reported | simonwillison.net ↗ | |
Standard TTS benchmarks OpenAI did not publish MOS (Mean Opinion Score) or WER metrics for gpt-4o-mini-tts. The announcement focuses on steerability and cost rather than quantitative voice quality benchmarks. | — | Not publicly disclosed | — |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/audio/speech -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "openai/gpt-4o-mini-tts",
"input": "Welcome to TheRouter.",
"voice": "alloy"
}'GPT-4o Mini TTS uses the standard OpenAI speech endpoint. Call it via TheRouter's unified API — the model supports synchronous and streaming modes with steerable instructions for tone control.
curl https://api.therouter.ai/v1/audio/speech \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-4o-mini-tts",
"input": "Welcome to TheRouter — your unified API for all AI models.",
"voice": "nova",
"instructions": "Speak warmly and professionally, like a friendly customer service agent.",
"response_format": "mp3"
}' \
--output speech.mp3OpenAI announced gpt-4o-mini-tts, a cost-efficient steerable text-to-speech model at $0.015/min, alongside new gpt-4o-transcribe and gpt-4o-mini-transcribe speech-to-text models. The model introduces natural-language instructions for tone, pace, and emotion control — a first for OpenAI TTS.
The model ships with 11 core preset voices: Alloy, Ash, Ballad, Coral, Echo, Fable, Nova, Onyx, Sage, Shimmer, and Verse. Additional voices Marin and Cedar were added in later updates. All voices are synthetic presets — the model does NOT support custom voice creation or voice cloning. You can preview all voices on openai.fm.
The instructions parameter accepts natural language guidance on emotion (excited, calm, urgent), pace (faster, slower), pronunciation emphasis, accent, and character direction. For example: 'Speak with urgency and a serious news-anchor tone.' The model blends these instructions with the base voice characteristics. Note: because instructions and text share the same input channel, there is a non-zero risk that part of the input text could be accidentally interpreted as instructions — a known limitation of LLM-based TTS.
The model supports 50+ languages, covering most major global languages. Voice characteristics are retained across languages — the same voice can speak multiple languages with consistent identity. For best results with non-English languages, OpenAI recommends selecting a voice that matches the target language's tonality.
Pricing is token-based: $0.60 per million input (text) tokens and $12 per million audio output tokens. At typical speech rates (~1,500 audio tokens per minute), this works out to approximately $0.015 per minute of generated speech. Through TheRouter, the pricing follows the same structure with a per-request fee of $0.015, input at $0.60/MTok, and audio output at $12/MTok — matching OpenAI's direct pricing with no markup.
Yes — gpt-4o-mini-tts is fully available on TheRouter at api.therouter.ai/v1. Use the standard OpenAI SDK with baseURL=https://api.therouter.ai/v1 and model=openai/gpt-4o-mini-tts. All OpenAI TTS features are supported: the instructions parameter, streaming, all voices and response formats. No additional configuration needed — just swap the base URL and model name.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | openai.com ↗ | 2026-05-28 | verified |
| Base architecture | openai.com ↗ | 2026-05-28 | verified |
| Endpoint | developers.openai.com ↗ | 2026-05-28 | verified |
| Voices | docs.aimlapi.com ↗ | 2026-05-28 | verified |
| Supported languages | tokenmix.ai ↗ | 2026-05-28 | verified |
| Context window | tokenmix.ai ↗ | 2026-05-28 | to verify |
| Output formats | developers.openai.com ↗ | 2026-05-28 | verified |
| Steerability | simonwillison.net ↗ | 2026-05-28 | verified |
| Steerability benchmark (Simon Willison) | simonwillison.net ↗ | 2026-05-28 | verified |
| Standard TTS benchmarks | openai.com ↗ | 2026-05-28 | unknown |
| OpenAI launches gpt-4o-mini-tts alongside new speech-to-text models | openai.com ↗ | 2026-05-28 | verified |
| What voices are available with gpt-4o-mini-tts? | docs.aimlapi.com ↗ | 2026-05-28 | to verify |
| How does the instructions parameter work? | simonwillison.net ↗ | 2026-05-28 | to verify |
| What languages does gpt-4o-mini-tts support? | tokenmix.ai ↗ | 2026-05-28 | to verify |
| How does pricing work exactly? Token-based vs per-minute? | tokenmix.ai ↗ | 2026-05-28 | to verify |
| Can I use gpt-4o-mini-tts through TheRouter? | api.therouter.ai ↗ | 2026-05-28 | to verify |