Provides emergency audio input
AI Models & Assistants / Knowledge library
GPT-Realtime-2
Reasoning model with tool use
Why learn GPT-Realtime-2?
Offers the documented OpenAI model capabilities through hosted.
The problem it solves
Provides model inference for supported text / audio / image inputs.
Where it came from
GPT-Realtime-2: Reasoning model with tool use Published aliases: gpt-realtime-2. Input: text, audio, image. 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-Realtime-2 does.
- Identify a use case for GPT-Realtime-2.
- Describe a limitation of GPT-Realtime-2.
- Name a compatible tool for GPT-Realtime-2.
- Compare GPT-Realtime-2 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
Automated Emergency Dispatch Coordination
Emergency medical dispatch centers managing high-volume, time-sensitive triage calls.
The business problem
Dispatchers struggle to maintain situational awareness while simultaneously querying multiple internal databases to verify patient history and resource availability, leading to cognitive overload and potential delays in critical information retrieval.
How this topic solves it
By utilizing the gpt-realtime-2 model, dispatch centers can deploy a voice-agent that uses its reasoning capabilities to interpret complex, non-linear emergency descriptions while concurrently executing tool calls to retrieve patient records and dispatch units in real-time.
- Configure the gpt-realtime-2 session via the v1/realtime endpoint to enable bi-directional audio streaming.
- Define specific function calling schemas for the agent to query external patient databases and unit availability systems.
- Set the reasoning effort parameter to balance the need for deep analytical processing of emergency symptoms against the requirement for low-latency voice responses.
- Implement the realtime event loop to handle incoming audio streams and outgoing tool call requests.
- Monitor the reasoning token usage to ensure the agent maintains a fluid, conversational pace during high-stress triage scenarios.
The flow shows how the model processes voice input, reasons through data, and uses tools to assist the dispatcher.
Processes audio and performs reasoning
Retrieves patient history
Receives synthesized summary
- Caller → Streams audio → gpt-realtime-2
- gpt-realtime-2 → Executes tool call → Database Tool
- Database Tool → Returns data → gpt-realtime-2
- gpt-realtime-2 → Provides voice summary → Dispatcher
Expected business outcome
The system achieves a more cohesive triage process where the agent acts as a cognitive assistant, allowing human dispatchers to focus on high-level decision-making while the model handles the technical retrieval and synthesis of data.
Limitations and checks
A primary trade-off is the latency introduced by higher reasoning effort settings; developers should monitor the time-to-first-token and total response duration to ensure the conversational flow remains acceptable for emergency contexts.