VideoFDB - Perception (Audio Only) - TOR-Alignment: leaderboard
Metric: TOR-Alignment (%; share of the timing-relevant held-out perception test clips on which the agent's floor-taking matches the expected timing class: stay silent, keep speaking, yield within 1.5 s, take a smooth handoff or produce a backchannel; measured deterministically from word-level timestamps of the agent's audio, streamed through a realtime harness with the video stream disabled (audio only); the median latency printed in the same cell is not read). Source: arxiv.org. Saturation forecast: Around December 2026. 7 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | MiniCPM-o-4.5 | 72 |
Interactive version: theaggregate.ai/benchmark?slug=videofdb-perception-audio-only-tor-alignment · How It Works · Data refreshed daily, snapshot 2026-09-29.