Captures raw audio input
AI Models & Assistants / Knowledge library
GPT-Audio-1.5
The best voice model for audio in, audio out with Chat Completions.
Why learn GPT-Audio-1.5?
Offers the documented OpenAI model capabilities through hosted.
The problem it solves
Provides model inference for supported text / audio inputs.
Where it came from
GPT-Audio-1.5: The best voice model for audio in, audio out with Chat Completions. Published aliases: gpt-audio-1.5. Input: text, audio. Output: text, audio. Checked 2026-10-13. Model capabilities are distinct from tools available in a particular agent.
Build your understanding
Work through these exercises in order. Try the task before opening a reference.
- Explain what GPT-Audio-1.5 does.
- Identify a use case for GPT-Audio-1.5.
- Describe a limitation of GPT-Audio-1.5.
- Name a compatible tool for GPT-Audio-1.5.
- Compare GPT-Audio-1.5 against an alternative.
Documentation and sources
- Primary source / model card (developers.openai.com)
Catalog research notes
- See the AI Workbench for source dates, pricing evidence and configuration details.
Connect the ideas
Built on
Real World Example
Multilingual Technical Support Voice Interface
Global consumer electronics manufacturers providing remote troubleshooting for non-English speaking customers.
The business problem
Traditional text-based support bots fail to capture the nuance of hardware issues described through ambient noise or tone, while standard speech-to-text pipelines introduce latency and lose emotional context during complex troubleshooting.
How this topic solves it
GPT-Audio-1.5 enables a native audio-in, audio-out loop via the Chat Completions API. By processing raw audio directly, the model interprets the user's vocalized frustration and technical descriptions without the intermediate step of transcription, allowing for immediate, empathetic, and accurate verbal guidance.
- Capture raw audio stream from the customer's device microphone.
- Encode the audio input into the required format for the Chat Completions API request.
- Send the payload to the gpt-audio-1.5 model endpoint with appropriate system instructions.
- Receive the audio output stream directly from the model response.
- Play the audio response back to the user to maintain a seamless, conversational troubleshooting experience.
This diagram illustrates the data flow from the customer to the model and back.
Sends audio to Chat Completions
Processes audio and generates response
Delivers voice guidance to user
- Customer Device → stream audio → API Request
- API Request → POST /v1/chat/completions → GPT-Audio-1.5
- GPT-Audio-1.5 → return audio content → Audio Output
Implementation example
This snippet demonstrates the core integration where raw audio bytes are passed as the user message content to the Chat Completions endpoint.
client.chat.completions.create(model='gpt-audio-1.5', messages=[{'role': 'user', 'content': audio_bytes}]) # Sends raw audio directly to the modelExpected business outcome
Improved customer satisfaction through reduced friction in technical support interactions and more accurate diagnosis of hardware problems based on auditory cues.
Limitations and checks
The model operates via the Chat Completions API rather than a persistent real-time session, which can introduce slight delays between turns; developers should monitor the time-to-first-byte for audio segments to ensure the conversation remains fluid.