Overview
GPT-Realtime-1.5 is a speech-to-speech model for low-latency voice agents and customer support, available through the Realtime API. OpenAI describes it as a fast, non-reasoning model. Compared with earlier Realtime preview models, OpenAI reports improved instruction following, tool calling and voice quality. (developers.openai.com)
Capabilities and uses
- Accepts text and audio input and produces text and audio output; it also accepts image input.
- Supports function calling, making it suitable for voice agents that need to interact with application tools.
- Has a 32,000-token context window, a maximum output of 4,096 tokens, and a knowledge cutoff of September 30, 2024. (developers.openai.com)
Limitations
- Video and image output are not supported.
- Structured outputs, fine-tuning and predicted outputs are not supported.
- It is a non-reasoning model; applications requiring stronger deliberative reasoning may need a different model. (developers.openai.com)
Release
Released to the Realtime API on February 23, 2026. The announcement separately identifies gpt-audio-1.5 as the model released to Chat Completions. (developers.openai.com)
Pricing
OpenAI API pricing in USD per 1 million tokens, verified November 16, 2026. Cached-input rates apply to eligible cached input; input and output are billed separately. (developers.openai.com)
| Modality | Input | Cached input | Output |
|---|---|---|---|
| Text | $4.00 | $0.40 | $16.00 |
| Audio | $32.00 | $0.40 | $64.00 |
| Image | $5.00 | $0.50 | Not supported |