GPT Audio
GPT Audio is a multimodal model made by OpenAI, added to the catalog on 19 Jan 2026. It has a context window of 128,000 tokens and costs $2.50 per 1M input tokens and $10.00 per 1M output tokens. It accepts text and audio and produces text and audio. Typical uses: generate audio from text with high fidelity, transcribe audio to text accurately and create engaging voiceovers for content.
Price checked ยท source: OpenRouter
What GPT Audio is
OpenAI's GPT Audio model, released on January 19, 2026, is the first audio model from OpenAI that enables seamless interaction between text and audio. With a context length of 128,000 tokens, it supports both text and audio input/output modalities, making it versatile for various applications. The model features an upgraded decoder that produces more natural-sounding voices while ensuring better voice consistency. This capability allows users to generate high-quality audio outputs from text and vice versa, enhancing the overall user experience in audio processing tasks.
Key features
- Upgraded decoder for natural-sounding voices
- Supports text and audio input/output
- Context length of 128,000 tokens
- Enhanced voice consistency
- Versatile for various audio applications
Typical uses
- Generate audio from text with high fidelity
- Transcribe audio to text accurately
- Create engaging voiceovers for content
- Develop interactive audio applications
GPT Audio pricing
US dollars, list price through OpenRouter.
| Item | Standard |
|---|---|
| Input, per 1M tokens | $2.50 |
| Output, per 1M tokens | $10.00 |
No price change since it was added on 19 Jan 2026.
| What a task costs | Cost |
|---|---|
| Answer a short question500 in / 300 out | $0.0043 |
| Summarise a 50-page PDF25,000 in / 800 out | $0.071 |
| Review a 2,000-line source file30,000 in / 2,000 out | $0.095 |
| Read a 300-page contract set150,000 in / 1,500 out | $0.390 |
GPT Audio specifications
| Model id | openai/gpt-audio |
|---|---|
| Version | gpt-audio |
| Context window | 128,000 tokensAbout 96,000 English words |
| Max output | 16,384 tokens |
| Knowledge cutoff | Not published |
| Added to catalog | 19 Jan 2026 |
| Input types | text, audio |
|---|---|
| Output types | text, audio |
| Tool calling | Yes |
| Structured outputs | Yes |
| Reasoning effort control | No |
| Content moderation | Applied by provider |
Sample outputs
The same prompts are given to every model, and the unedited output is shown with the date it was produced. Upend does not publish leaderboard scores it did not run.
No output has been recorded for GPT Audio yet.