GPT-Realtime-2 accessible via WebRTC API but still absent from ChatGPT app, prompting developer workarounds
OpenAI's GPT-Realtime-2 — billed as "the first voice model with GPT-5-class reasoning" with a Sep 30, 2024 knowledge cutoff — has been available through the company's WebRTC Realtime API for weeks, but has yet to appear in the consumer ChatGPT app or iPhone client. The gap is wide enough that developers are routing around it.
Simon Willison updated his open-source OpenAI-WebRTC playground to expose model selection, letting users pick GPT-Realtime-2 directly and paste in arbitrary document context for browser-based audio conversations. The episode highlights a recurring friction point: OpenAI's API-tier and product-tier model rollouts running on noticeably different schedules, leaving developers to build their own front-ends to access capabilities that paying ChatGPT subscribers cannot yet reach.