Source of live audio input
AI Models & Assistants / Knowledge library
GPT-Realtime-Translate
Streaming speech-to-speech translation model
Why learn GPT-Realtime-Translate?
Offers the documented OpenAI model capabilities through hosted.
The problem it solves
Provides model inference for supported audio inputs.
Where it came from
GPT-Realtime-Translate: Streaming speech-to-speech translation model Published aliases: gpt-realtime-translate. Input: audio. Output: audio, text. 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-Realtime-Translate does.
- Identify a use case for GPT-Realtime-Translate.
- Describe a limitation of GPT-Realtime-Translate.
- Name a compatible tool for GPT-Realtime-Translate.
- Compare GPT-Realtime-Translate 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
Global Emergency Response Coordination
International disaster relief organizations managing field operations in regions where local dialects and languages shift rapidly.
The business problem
Traditional translation services introduce significant latency, causing critical delays in relaying life-saving instructions from command centers to local field teams during time-sensitive emergency operations.
How this topic solves it
The GPT-Realtime-Translate model provides a streaming speech-to-speech translation pipeline that processes incoming audio from command centers and outputs translated audio and transcript deltas simultaneously, allowing field teams to receive instructions in their native language while the source audio is still being transmitted.
- Establish a persistent connection to the v1/realtime/translations endpoint.
- Stream raw audio packets from the command center microphone directly to the model input.
- Receive real-time translated audio chunks and corresponding transcript deltas as they are generated.
- Inject the translated audio stream into the field team's communication devices for immediate playback.
- Monitor the streaming session to ensure the audio duration remains within the established rate limits for the organization's usage tier.
The flow of audio data from the command center through the translation model to the field team.
Streaming translation model
Receiver of translated audio
- Command Center → Streams raw audio → v1/realtime/translations
- v1/realtime/translations → Outputs translated audio and transcript deltas → Field Team
Expected business outcome
Field teams receive near-instantaneous, actionable instructions in their local language, significantly reducing the cognitive load and communication lag during high-stress emergency coordination.
Limitations and checks
The model operates as a streaming speech-to-speech engine, which means it does not support standard chat completion or batch processing workflows; developers must verify compatibility by ensuring their infrastructure is configured specifically for the v1/realtime/translations endpoint.