Simon Willison has updated his WebRTC audio session tool, originally built in December 2024 to experiment with OpenAI's real-time audio interaction capabilities. The updated tool now supports GPT-Realtime-2, a model OpenAI introduced last month. This new model is described by the company as having GPT-5-class reasoning and includes a knowledge cut-off date of September 30, 2024.

In addition to the model update, the tool now allows users to paste significant amounts of document context. This feature enables users to conduct spoken conversations in a browser about specific content they provide. Willison developed these enhancements after waiting for the new model to appear in the official ChatGPT iPhone app, which has not yet occurred. The tool leverages the WebRTC (a standard for real-time browser-based communication) API to facilitate these voice-based interactions.

Simon Willison has updated his WebRTC audio session tool, originally built in December 2024 to experiment with OpenAI's real-time audio interaction capabilities. The updated tool now supports GPT-Realtime-2, a model OpenAI introduced last month. This new model is described by the company as having GPT-5-class reasoning and includes a knowledge cut-off date of September 30, 2024.

In addition to the model update, the tool now allows users to paste significant amounts of document context. This feature enables users to conduct spoken conversations in a browser about specific content they provide. Willison developed these enhancements after waiting for the new model to appear in the official ChatGPT iPhone app, which has not yet occurred. The tool leverages the WebRTC (a standard for real-time browser-based communication) API to facilitate these voice-based interactions.