OpenAI: GPT-Realtime-1.5 API
GPT-Realtime-1.5 is OpenAI's flagship low-latency multimodal model, specifically engineered for high-stakes voice AI and real-time interactive systems. It features a natively multimodal "Speech-to-Speech" architecture that bypasses traditional ASR (Automatic Speech Recognition) and TTS (Text-to-Speech) pipelines, significantly reducing latency and preserving emotional nuances. Compared to its predecessor, version 1.5 offers a 7% boost in instruction following and a 10% improvement in alphanumeric transcription accuracy. It natively supports interleaved multimodal inputs, allowing users to stream audio, images, and text simultaneously for complex tasks like "seeing" a problem via a camera while discussing it in real-time.
- Context window: 32,000 tokens
- Max output: 4,096 tokens
- Input: text, audio, image
- Output: text, audio
- Tool calling: Supported
- File input: Supported
- Released: 2026-02-22
- Knowledge cutoff: 2025-08-31
Frequently Asked Questions
What is the context window of GPT-Realtime-1.5?
GPT-Realtime-1.5 supports a context window of up to 32,000 tokens.
Does GPT-Realtime-1.5 support function calling?
Yes. GPT-Realtime-1.5 supports tool / function calling.
What is the knowledge cutoff of GPT-Realtime-1.5?
The knowledge cutoff of GPT-Realtime-1.5 is 2025-08-31.