ByteDance launched SeedRealtime, a “full-duplex multimodal model” that processes audio, video, and text simultaneously in real time, on April 9 via its internal AI research group Seed. The model is available to users of the Doubao app, which boasts 155 million weekly active users, with daily active users surpassing 200 million shortly after the Lunar New Year in 2026. SeedRealtime reportedly delivers a 12% improvement in conversational fluency compared to its predecessors, attributed to advanced listening techniques and interference suppression. Doubao powers interactions across ByteDance products including Coze, Jimeng, and Volcano Engine. The Seed team, established in 2023 with research centers in China, Singapore, and the United States, released Seed 2.0 in February 2026, with Seedance 2.5 expected in late July 2026—three major model launches in roughly five months. ByteDance has made zero moves to tokenize its AI products or integrate with blockchain protocols, keeping its AI stack centralized and off-chain. The article notes that companies shipping AI products at massive scale are not the ones building on blockchain rails, highlighting a tension with decentralized AI projects. OpenAI and Google’s Gemini are also pushing similar real-time audio-visual capabilities.