OpenAIGPT Realtime

GPT-Realtime-1.5

GPT-Realtime-1.5 supports interactive voice agents with text, audio and image input. It returns text or speech and can call application tools.

The older session limits still matter

Its window is 32,000 tokens and its output limit is 4,096 tokens. Prompt caching is supported. Audio input and output use separate rates from text, so text-token prices alone do not describe the cost of a call.GPT-Realtime-1.5 official model documentation

Plan for a bounded conversation history

Practical guidance. For an existing phone assistant, inspect how long calls retain instructions and recent tool results within the session budget. Compare a newer model on the same call flow before migration. Preserve important confirmed facts when the application shortens conversation history.

Specifications & access

GPT-Realtime-1.5 specifications
DetailValue and conditions
AccessAPI
Model identifiersgpt-realtime-1.5
Context window32,000 tokens

Published model context; input and output share this budget

GPT-Realtime-1.5 official model documentation
Maximum output4,096 tokens

Published maximum output; bounded by the remaining context

GPT-Realtime-1.5 official model documentation
Input and outputtext, audio, image in; text, audio outGPT-Realtime-1.5 official model documentation
API routesRealtime (v1/realtime)GPT-Realtime-1.5 official model documentation

Related models

  • GPT-Realtime-2.1

    OpenAI · GPT Realtime

    GPT-Realtime-2.1 combines conversational audio with reasoning and tool use.

    Launch date not confirmedAPI
  • GPT-Realtime-2

    OpenAI · GPT Realtime

    GPT-Realtime-2 is a speech-to-speech model with configurable reasoning and tools.

    Launch date not confirmedAPI
  • GPT-Realtime-2.1 Mini is a distilled voice-reasoning model for lower-cost interactive audio.

    Launch date not confirmedAPI
  • GPT-Realtime-Translate converts incoming speech into translated speech and text while the source audio is still arriving.

    Launch date not confirmedAPI