VideoFDB - Perception (Audio Only) - Conversational Flow: leaderboard

The conversational flow rubric of VideoFDB (Mazumdar et al., 2026) rates whether responses are timed correctly around the nonverbal cue, such as staying silent through a pause or yielding to an interruption. 105 held-out test clips of seven nonverbal conversational dynamics, such as pauses, backchannels, laughter and interruptions, are streamed in real time to a full-duplex speech agent with the video stream disabled, so it hears but does not see the user, and a gpt-4o judge scores each response from 0 to 5; higher is better.

Metric: gpt-4o judge score (0-5; the 105 held-out test clips of the seven nonverbal conversational dynamics scored for perception (237 clips of 11 dynamics in the full benchmark), streamed in real time; the judge reads the transcribed, time-stamped agent response with the clip annotations; timing of responses around the nonverbal cue, on the dynamics this rubric covers; the same clips with the video stream disabled). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 7 models tracked.

Top models

Interactive version: theaggregate.ai/benchmark?slug=videofdb-perception-audio-only-conversational-flow · How It Works · Data refreshed daily, snapshot 2026-11-12.