ByteDance's Seed team unveils real-time audio-visual AI model

ByteDance's Seed team introduced SeedRealtime, a model that fuses audio, video, and text into one architecture and interacts over continuous multimodal streams rather than turn by turn. The team calls it a step toward omni-modal interaction, citing joint audio-visual understanding as one of its capabilities.

Why it matters: Full-duplex, always-on multimodal models are a different architecture bet than the turn-based chat models most labs ship today, aimed at applications like live video assistants and real-time translation. It's another sign of ByteDance pushing to compete at the model layer, alongside its reported 10-trillion-parameter model aimed at Anthropic.