This post was written on Dec 17, 2025.
Models/pricing/policies may have changed. Check the latest openai posts.
OpenAI Realtime API & WebRTC: Conquering Voice AI Latency
WebRTC support has been added to the OpenAI Realtime API. Now, ultra-low latency voice conversations are possible in web browsers and mobile apps. We analyze the technical implications.
Did you find the subtle delay in voice conversations with AI frustrating? (Problem) In December 2025, OpenAI officially started supporting WebRTC connection methods in the Realtime API. (Solution) This drastically reduces network latency compared to existing WebSocket methods, enabling instant reaction speeds (Low Latency) akin to talking to a human. (Evidence)
Changes Brought by WebRTC
1. Browser Native Support
You can open audio/video channels directly with AI models through WebRTC, a browser standard technology, without separate complex audio stream processing. Development difficulty decreases while quality increases.
2. Seamless Conversation (Interruptibility)
The reaction speed at which the AI detects and stops speaking when a user interrupts or cuts in (Barge-in) has dramatically improved. Natural 'back-and-forth' becomes possible.
3. Multimodal Streaming
Not just voice, but video feeds can also be transmitted in real-time. For example, it becomes possible to implement a service where you point a camera at a math problem and talk to an AI tutor in real-time.
gpt-realtime-mini Model
The simultaneously released gpt-realtime-mini model maximizes cost-efficiency, opening the way for large-scale B2C services to adopt voice AI without burden.
Get updates
A weekly digest of what actually matters.
Found an issue? Report a correction so we can review and update the post.