OpenAI WebRTC Audio Session, now with document context

Simon Willison's Blog Tools
openai webrtc audio realtime document-context gpt-realtime-2 playground

Summary

Simon Willison updates his OpenAI WebRTC Audio Session tool to support the new GPT-Realtime-2 model and adds document context for conversational audio discussions.

No content available
Original Article
View Cached Full Text

Cached at: 06/13/26, 03:30 PM

# OpenAI WebRTC Audio Session, now with document context Source: [https://simonwillison.net/2026/Jun/12/openai-webrtc/](https://simonwillison.net/2026/Jun/12/openai-webrtc/) 12th June 2026 \- Link Blog **[OpenAI WebRTC Audio Session, now with document context](https://tools.simonwillison.net/openai-webrtc)**\. I built the first version of this tool[in December 2024](https://simonwillison.net/2024/Dec/17/openai-webrtc/)to try out the then\-new OpenAI WebRTC API for interacting with their realtime audio models\. Last month OpenAI[introduced a brand new model](https://openai.com/index/advancing-voice-intelligence-with-new-models-in-the-api/)to that API called[GPT‑Realtime‑2](https://developers.openai.com/api/docs/models/gpt-realtime-2), which they promoted as "our first voice model with GPT‑5‑class reasoning" \- with a Sep 30, 2024 knowledge cut\-off\. I've been waiting for that model to show up in the ChatGPT iPhone app but it still hasn't, so I revisited my old playground\. You can now pick the better model, and you can also paste in a big chunk of document context so you can have as audio conversation in your browser about whatever information you think would be useful to explore in a conversational way\. ![Screenshot of a web interface titled "OpenAI WebRTC Audio Session" with a gray status dot. Form fields: "OpenAI API Token" showing a masked password of dots, "Voice" dropdown set to "Coral", "Model" dropdown set to "gpt-realtime-2". A collapsible section labeled "▼ Document context (optional — paste text to talk about)" with bold instruction "Paste a document here before starting the session and the model will be able to discuss it with you" above a textarea containing a pasted Markdown document about whether DuckDB can run untrusted SQL as safely as Datasette runs SQLite. Below are a blue "Start Session" button and a gray disabled "Mute Mic" button, then a green success message "Session established successfully!" At the bottom, a dark panel headed "Last transcript" reads: "DuckDB can be made about as safe as SQLite for running untrusted SELECT queries, but only if you lock it down properly. Using read only true by itself is not enough, because SQL can still" (text cut off).](https://static.simonwillison.net/static/2026/openai-webrtc-document-context.jpg)

Similar Articles

@danshipper: Lfgggg!!!

X AI KOLs Timeline

A design lead from OpenAI is confirmed to speak at the Thesis conference on work and AI, discussing evolving work practices such as content designers submitting PRs and PMs bringing prototypes to design reviews.

GPT-5.6 Sol Pricing Cut by 50%

Hacker News Top

OpenAI has announced a 50% reduction in the API pricing for its GPT-5.6 Sol model, making it more cost-effective for developers. The update is highlighted on OpenRouter's platform with links to benchmarks and documentation.