Benchmarks
Benchmarks
LVBench
Visionunit % · normalized over [0, 92]LVBench — long-video understanding over hour-scale footage, testing retrieval and reasoning across extended temporal context.
#ModelSourceScoreNormalized
Score distribution
6 tracked results across the normalization window
092
Score vs. parameters
Open-weights models, log-x params
Only one open-weights model with a disclosed parameter count — not enough to plot a trend.