Provides voice input
AI Models & Assistants / Knowledge library
GPT-4o Audio
GPT-4o models capable of audio inputs and outputs
Why learn GPT-4o Audio?
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-4o Audio: GPT-4o models capable of audio inputs and outputs Published aliases: gpt-4o-audio-preview-2025-06-03, gpt-4o-audio-preview-2024-12-17, gpt-4o-audio-preview-2024-10-01. 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-4o Audio does.
- Identify a use case for GPT-4o Audio.
- Describe a limitation of GPT-4o Audio.
- Name a compatible tool for GPT-4o Audio.
- Compare GPT-4o Audio 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 technical support centers providing hardware troubleshooting for non-native speakers.
The business problem
Support agents struggle to interpret complex technical descriptions from users with thick accents or limited vocabulary, leading to high ticket resolution times and user frustration.
How this topic solves it
Implementing the gpt-4o-audio-preview model allows the system to ingest raw audio input directly, bypassing the latency and potential data loss of intermediate speech-to-text transcription layers. By processing the user's voice directly, the model captures nuanced prosody and tone, enabling more accurate intent recognition and providing immediate, natural-sounding audio responses that guide the user through troubleshooting steps in their preferred language.
- Capture the user's audio stream via the client-side interface.
- Format the audio data into a base64-encoded string for the Chat Completions API payload.
- Send the request to the gpt-4o-audio-preview model specifying audio as both input and output modality.
- Receive the response containing the generated audio content.
- Stream the returned audio output back to the user's device for immediate playback.
This diagram illustrates the flow of audio data from the user through the GPT-4o Audio model and back.
Encodes audio to base64
Processes gpt-4o-audio-preview
Decodes and plays audio
- User → records voice → Client App
- Client App → sends audio input → Chat Completions API
- Chat Completions API → returns audio output → Audio Response
Implementation example
This snippet demonstrates sending a base64-encoded audio input to the model and requesting an audio response, which is central to the Multilingual Technical Support Voice Interface.
client.chat.completions.create(
model="gpt-4o-audio-preview-2025-06-03",
messages=[{"role": "user", "content": "[base64_audio_data]"}],
modalities=["audio"] # Requesting audio output
)Expected business outcome
Improved user experience through reduced communication friction and faster, more empathetic technical guidance without the need for manual transcription.
Limitations and checks
The model operates under a strict 16,384 max output token limit, which may truncate long-form technical instructions; developers should monitor the finish_reason in the API response to check if the output was cut off.