Boson AI has released Higgs Realtime Duplex, a 45 billion parameter Mixture-of-Experts model capable of simultaneous audio and video processing, alongside text generation and tool use. This full-duplex system processes information in 160ms chunks, allowing for continuous background work such as thinking, tool delegation, and response generation even while speaking. The model can be zero-shot configured with reference audio and images for voice and face, and has demonstrated strong performance across several benchmarks, though it does not offer open weights. AI
IMPACT Sets a new bar for real-time multimodal interaction, potentially accelerating the development of more natural and responsive AI agents.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →