GPT Audio
voice and audio modelOpenAI
The gpt-audio model is OpenAI's first generally available audio model. The new snapshot features an upgraded decoder for more natural sounding voices and maintains better voice consistency. Audio is priced...
Key facts
| Made by | OpenAI |
|---|---|
| Type | voice and audio model |
| Released | January 19, 2026 |
| Reads at once | 128,000 tokens |
| Price (API, per million tokens) | $2.5 in · $10 out |
| Input → output | text, audio → text, audio |
Questions about GPT Audio
Who makes GPT Audio?
GPT Audio is made by OpenAI.
When was GPT Audio released?
We first saw GPT Audio listed on January 19, 2026.
How much does GPT Audio cost?
Through OpenRouter it costs $2.5 per million input tokens and $10 per million output tokens. Prices change, so check the provider before you build on it.
How much text can GPT Audio read at once?
Up to 128,000 tokens, roughly 96K words.
More from OpenAI
Whisper-SmallOpenAI · voice and audio model
Sep 26, 2022released
Whisper-Large-V3OpenAI · voice and audio model
Nov 7, 2023released
whisperOpenAI · voice and audio model
Sep 26, 2022released
Whisper-TinyOpenAI · voice and audio model
Sep 26, 2022released
Openai-Agents-Pythonopenai · developer tool
Mar 11, 2025released
Whisper-MediumOpenAI · voice and audio model
Sep 26, 2022released
Other ai voice and audio
Wav2vec2-Large-Xlsr-53-JapaneseJonatasgrosman · voice and audio model
Mar 2, 2022released
Whisperkit-CoremlArgmaxinc · voice and audio model
Feb 28, 2024released
Speaker-Diarization-3.1pyannote · voice and audio model
Nov 16, 2023released
XTTS-v2Coqui · voice and audio model
Oct 31, 2023released
Speaker-Diarization-Community-1pyannote · voice and audio model
Apr 15, 2025released
Faster-Whisper-SmallSYSTRAN · voice and audio model
Nov 23, 2023released
Wav2vec2-Large-Xlsr-53-RussianJonatasgrosman · voice and audio model
Mar 2, 2022released
Wav2vec2-Large-Xlsr-53-PortugueseJonatasgrosman · voice and audio model
Mar 2, 2022released
Modelcast collects this page automatically from public sources (OpenRouter). Facts are copied from those sources, not written by the maker. Not affiliated with OpenAI.