OpenAI Ships Realtime Video API: Stream and Analyze Live Video with GPT-5V

OpenAI has launched its long-awaited Realtime Video API, a WebRTC-based interface that allows developers to stream live video feeds directly into GPT-5V and receive structured JSON responses in near real time. The API supports up to 1080p30 input and returns object detections, scene descriptions, and free-form answers to natural-language queries with median latency under 400ms.

The release opens significant new use cases: accessibility apps that narrate surroundings for visually impaired users, manufacturing QA systems that inspect products on a conveyor belt, and robotics pipelines that replace expensive custom vision models with a single API call. Pricing is set at $0.02 per second of video processed, which the company notes is roughly comparable to running a 4K LiDAR sensor in cost-per-insight terms.

Early developer previews have already shown impressive demos including a real-time sign-language interpreter and an autonomous drone navigation system that uses GPT-5V as its sole perception backbone. OpenAI's documentation confirms the API is compatible with the existing Assistants SDK, meaning teams can layer in memory, tool-calling, and function execution on top of the video stream.