OpenAI adds GPT-Realtime-2 + realtime translation + realtime transcription models to the API (May 2026)

OpenAI announced new realtime voice models for developers:

• GPT-Realtime-2: OpenAI’s “most capable realtime voice model,” designed for speech-to-speech interactions with configurable reasoning effort. (OpenAI model docs: https://developers.openai.com/api/docs/models/gpt-realtime-2)
• gpt-realtime-translate: a streaming speech-to-speech translation model that returns translated audio plus transcript deltas while source audio is still arriving. (OpenAI model docs: https://developers.openai.com/api/docs/models/gpt-realtime-translate)
• gpt-realtime-whisper: a streaming speech-to-text model for low-latency transcript deltas from live audio. (OpenAI model docs: https://developers.openai.com/api/docs/models/gpt-realtime-whisper)

OpenAI’s developer announcement frames this as an upgrade to “voice agents,” highlighting GPT-Realtime-2 plus dedicated translation and transcription models. (OpenAI Developers on X: https://x.com/OpenAIDevs/status/2052440907933474954)

GPT-Realtime-2: higher capability for voice agents

OpenAI’s docs describe gpt-realtime-2 as a reasoning model for realtime voice interactions, supporting text + audio I/O and image input. (https://developers.openai.com/api/docs/models/gpt-realtime-2)

Notable specs shown in the docs:

• Context window: 128,000 tokens
• Max output tokens: 32,000 tokens
• Endpoints listed: includes v1/realtime plus the standard API endpoints.

(https://developers.openai.com/api/docs/models/gpt-realtime-2)

The docs also show token-based pricing for text/audio/image for this model. (https://developers.openai.com/api/docs/models/gpt-realtime-2)

gpt-realtime-translate: streaming speech-to-speech translation

OpenAI’s docs describe gpt-realtime-translate as a streaming speech-to-speech translation model for live multilingual audio experiences. (https://developers.openai.com/api/docs/models/gpt-realtime-translate)

OpenAI says it uses a dedicated realtime translation endpoint and streams translated audio and transcript deltas while source audio arrives. (https://developers.openai.com/api/docs/models/gpt-realtime-translate)

The docs show per-minute audio pricing for this model. (https://developers.openai.com/api/docs/models/gpt-realtime-translate)

gpt-realtime-whisper: low-latency transcription

OpenAI’s docs describe gpt-realtime-whisper as a streaming speech-to-text model for realtime transcription, designed for low-latency transcript deltas from live audio. (https://developers.openai.com/api/docs/models/gpt-realtime-whisper)

The docs show per-minute audio pricing and list the realtime transcription endpoint as v1/realtime/transcription_sessions. (https://developers.openai.com/api/docs/models/gpt-realtime-whisper)

Why this matters

If you’re building realtime voice experiences, OpenAI is signaling a clearer split between:

• a general-purpose voice agent model (gpt-realtime-2) and
• specialized “pipe” models for translation and transcription.

This can simplify production architectures where you need translation and captions alongside agent reasoning.

Sources

• OpenAI Developers announcement: https://x.com/OpenAIDevs/status/2052440907933474954
• OpenAI model docs (gpt-realtime-2): https://developers.openai.com/api/docs/models/gpt-realtime-2
• OpenAI model docs (gpt-realtime-translate): https://developers.openai.com/api/docs/models/gpt-realtime-translate
• OpenAI model docs (gpt-realtime-whisper): https://developers.openai.com/api/docs/models/gpt-realtime-whisper