ByteDance Seed Introduces SeedRealtime Multimodal Full-Duplex Model
ByteDance’s Seed team has presented SeedRealtime, a full-duplex large language model designed to process audio, video and text within one architecture. The system handles perception, interpretation, decisions and responses concurrently across continuous multimodal streams, rather than waiting for separate conversational turns. Seed describes this as progress toward omni-modal interaction, highlighting unified audio-visual understanding, proactive responses and more natural timing.
The demonstrations show the model associating voices with faces during a noisy dinner, speaking when a requested bronze screen stand appears at the Hebei Museum, and identifying a section of a quickly turned ResNet paper before reading its technical values. In an espresso demonstration, it detects a preparation mistake from the visual scene and recommends shortening extraction by 2 to 3 seconds. At Beijing Daxing Airport, it ignores unrelated conversation, recalls information that has left the screen and retrieves the baggage-carousel location online.
SeedRealtime is already operating in the Doubao app. However, ByteDance has not released a technical report, parameter count, open weights or an endpoint through Volcano Engine or BytePlus. The company’s human evaluation reported that pacing problems were reduced by half compared with cascaded systems, but provided no benchmark or latency figures.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.