Check the date
Voice products change quickly. Prefer current Help and model pages when a replay and the live contract differ.
Watch and build
Start with OpenAI's official GPT-Live replay, then use the developer demos and voice-agent examples to separate ChatGPT product behavior from the Realtime API you can build with today.
Last verified:
Featured official replay
OpenAI's July 8, 2026 replay demonstrates the GPT-Live conversation model, interruption behavior, visual answers, and the product launch context.
Primary product demonstrations and launch context from OpenAI.
Official launch replay for the GPT-Live-powered Voice experience.
Verified
Launch article with interactive examples for cascaded, turn-based, and full-duplex voice behavior.
Verified
Official archive that identifies the GPT-Live replay and other relevant product and Realtime sessions.
Verified
Use documented APIs and connection patterns rather than copying the ChatGPT interface.
Start here for the current low-latency audio and multimodal developer contract.
Verified
Architecture guidance for voice agents, tools, handoffs, latency, and production decisions.
Verified
Recommended browser and mobile connection path with ephemeral client authentication.
Verified
Official and community-oriented examples for seeing voice-agent patterns in a working project.
A Realtime API showcase that lets users talk to an interactive 3D solar system.
Verified
Open-source examples for supervisor agents, sequential handoffs, tools, and voice guardrails.
Verified
A React reference app for inspecting WebRTC sessions, events, audio, and function calling.
Verified
Read the dated rollout timeline and current product/API boundary.
Open guideSeparate visual answers and image attachments from live camera video and screen sharing.
Open guideChoose a model, transport, price tier, and production architecture.
Open guideVoice products change quickly. Prefer current Help and model pages when a replay and the live contract differ.
A ChatGPT feature demonstration does not imply that the same GPT-Live model or interface is available to developers.
Use OpenAI's docs and example repositories as a starting point, then add your own auth, consent, safety, and observability.