Overview
GPT-Realtime-2.1 is a realtime reasoning model for speech-to-speech interactions and voice-agent workflows. Compared with GPT-Realtime-2, OpenAI reports improvements in alphanumeric recognition, silence and noise handling, and interruption behavior. It supports instruction following, function calling, and configurable reasoning effort. (developers.openai.com)
Capabilities and limitations
- Text and audio are supported as both input and output; images are input-only. Video is not supported.
- The context window is 128,000 tokens, with a maximum of 32,000 output tokens. The knowledge cutoff is September 30, 2024.
- Higher reasoning effort can increase latency and output-token usage.
- The model page lists structured outputs, fine-tuning, predicted outputs, and its general streaming feature as unsupported. (developers.openai.com)
Release
Released on July 6, 2026, according to the official OpenAI API changelog. (developers.openai.com)
API pricing
OpenAI's published prices for gpt-realtime-2.1 are in USD per 1 million tokens. Cached input rates apply to eligible cached input; image output is not supported. (platform.openai.com)
| Modality | Input | Cached input | Output |
|---|---|---|---|
| Text | $4.00 | $0.40 | $24.00 |
| Audio | $32.00 | $0.40 | $64.00 |
| Image | $5.00 | $0.50 | Not supported |
Pricing source: OpenAI API pricing. Queried November 16, 2026. Tool use may incur separate charges. (platform.openai.com)