What changed
Hugging Face has announced the release of MOSS-Live, a full-duplex real-time interaction system designed to support continuous conversation with live visual understanding and background task execution. The system enables users to engage in natural, uninterrupted dialogue while simultaneously handling background tasks such as research, reasoning, and tool use.
What this means in practice
MOSS-Live supports full-duplex conversation, allowing users to listen and speak at the same time, handle genuine interruptions, and distinguish between pauses and requests to take the floor. It also features selective attention, enabling the system to differentiate speech directed at the assistant from background speech and nearby conversations. The system maintains continuity across conversational turns, retaining relevant context and returning to interrupted explanations when appropriate.
MOSS-Live integrates live visual understanding by incorporating camera or screen-sharing input into the conversation alongside speech and ambient sound. It allows for tool use without interrupting the conversation, dispatching research, reasoning, and tool use to the background while the foreground remains available for dialogue. The system supports concurrent listening and interpretation, responding to user-defined conditions such as corrections or reminders.
Limitations and availability
MOSS-Live is available for testing in the online playground and can be integrated into applications using the real-time API guide. The system has been evaluated against several benchmarks, including FDB v1, v1.5, v2, and v3, Big-Bench Audio, and τ-Voice. MOSS-Live achieved the best scores on all six interaction metrics, including avoiding premature takeovers, responding selectively to unrelated speech, continuing through irrelevant interruptions, and handling backchannels and vocal tics. It also achieved the highest tool-use success rate on FDB v3 and the second-highest score on τ-Voice for multi-turn task success.
The supplied announcement does not specify the exact hardware requirements or licensing conditions for MOSS-Live. The system is currently in preview and may not be available for all users. The full benchmark results and additional details are available in the provided source.
