GPT-Realtime-2

2026-05-07 · Publisher : OpenAI · Domain : Artificial Intelligence

OpenAI launches three audio models in the Realtime API: GPT-Realtime-2 for voice agents with reasoning, GPT-Realtime-Translate for live translation and GPT-Realtime-Whisper for transcription.

What changes

  • GPT-Realtime-2: speech-to-speech with configurable reasoning effort, better instruction following and more reliable tool calls.
  • GPT-Realtime-Translate: streaming translation, more than 70 input languages and 13 output languages, priced per minute.
  • GPT-Realtime-Whisper: low-latency streaming transcription for captions and live note-taking.
  • Compatible with EU data residency, with classifiers active during sessions.

OpenAI launches three audio models in the Realtime API: GPT-Realtime-2 for voice agents with reasoning, GPT-Realtime-Translate for live translation and GPT-Realtime-Whisper for transcription. ### What changes - GPT-Realtime-2: speech-to-speech with configurable reasoning effort, better instruction following and more reliable tool calls. - GPT-Realtime-Translate: streaming translation, more than 70 input languages and 13 output languages, priced per minute. - GPT-Realtime-Whisper: low-latency streaming transcription for captions and live note-taking. - Compatible with EU data residency, with classifiers active during sessions. ### Specifications - GPT-Realtime-2: context 128,000 tokens, maximum output 32,000 tokens, text, audio and image input; text and audio output. - Reasoning effort up to xhigh; knowledge cutoff: September 30, 2024. - Translate: context 16,000 tokens, output 2,000 tokens, dedicated endpoint v1/realtime/translations. - Identifiers: gpt-realtime-2, gpt-realtime-translate, gpt-realtime-whisper. ### Pricing | Model | Price | | --- | --- | | GPT-Realtime-2, audio | $32 input ($0.40 cached), $64 output per million tokens | | GPT-Realtime-2, text | $4 input ($0.40 cached), $24 output per million tokens | | GPT-Realtime-Translate | $0.034 per minute | | GPT-Realtime-Whisper | $0.017 per minute | ### Benchmarks | Evaluation | GPT-Realtime-2 vs GPT-Realtime-1.5 | | --- | --- | | Big Bench Audio (high effort) | +15.2% | | Audio MultiChallenge, instruction following (xhigh effort) | +13.8% | Scores published by the vendor. Baseline scores not detailed. ### Key takeaways Voice becomes a full agent channel: reasoning, tools and translation without chaining three models. The text price of GPT-Realtime-2 is identical to that of version 1.5.

Links

Tags : Audio, OpenAI, Realtime, Voix, Traduction, Transcription