Choose the evaluation track
Use Progressive for fixed-prompt long-horizon behavior and Interactive for runtime response, retention, and reuse.
Streaming systems, evaluated while they run.
Official metric-wise results for streaming audio-video generation across progressive and interactive evaluation tracks.
◆ Scores expose change over time—not only the final output.
Official results
Choose a metric below to rank systems. Metrics use different scales and are therefore reported separately.
Quality, alignment, continuity, and efficiency metrics applied across both tracks.
Metric-wise ranking
Perceptual quality and semantic or temporal correspondence between audio and video.
Ranked by Visual AestheticsHigher is better
The leaderboard data could not be loaded.
Reading the ranks. Click any metric header to rank by that metric. Published three-decimal ties use competition ranking (for example, 1, 1, 3). PVRL and PARL are conditional on successful target achievement and should be read with PVUAR and PAUAR.
HDF is evaluated only when its required source state was established. NBC, FPS, and TTFC are unavailable for PixVerse R1, HappyOyster, and Odyssey-2 because their web interfaces do not expose the required traces.
Metric guide
The cards update with the selected leaderboard category. Choose a card to rank the table by that metric.
How to read StreamAV-Bench
Use Progressive for fixed-prompt long-horizon behavior and Interactive for runtime response, retention, and reuse.
High quality or fulfillment can coexist with temporal drift. Read absolute scores alongside their drift counterparts.
Low conditional latency does not guarantee reliable updates. Read response latency together with achievement rate.