SeedRealtime
Native full-duplex audio-video model that watches, listens, and speaks at once
About
SeedRealtime is ByteDance Seed team's native audio-visual full-duplex model, launched August 2026. Instead of chaining ASR, vision, language, and TTS modules, it fuses audio, video, and text in one end-to-end architecture and runs perception, understanding, decision-making, and expression in parallel. Turn-taking happens inside the model rather than in an external voice-activity detector, which cuts the latency and information loss of cascaded pipelines. ByteDance claims joint audio-visual understanding, proactive interaction, and natural conversational timing. It is currently live inside the Doubao consumer app for real-time video interaction, AI assistants, and customer service scenarios. As of now ByteDance has published no technical report, no parameter count, no open weights, and no public API endpoint, so third-party developers cannot integrate it; it stands as a reference architecture and a moved goalpost for real-time multimodal products rather than a usable model.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
None published
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Launch Aug 2026 verified via TechNode and MarkTechPost; No technical report, params, weights, or API; honestly flagged as app-only; Thin independent verification, scored accordingly.