What it measures

Multi-task long-video understanding benchmarks evaluating needle-in-a-video retrieval, plot reasoning, and subtitle-visual alignment up to 60 minutes.

Reading the measurement

Study what information the model receives and which inference the answer requires. A question may mix recall, visual perception, deduction and search. Use task breakdowns to find the mechanism behind a headline gain.

Published evaluation

Test-set overall accuracy from the official paper. The validation total and duration/task breakdowns are distinct columns.

Where this fits in an evaluation suite

Ask what the task actually requires: recall, deduction, search, or a mixture. Separate closed-book reasoning from retrieval and tool-assisted problem solving.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

Test-set overall accuracy from the official paper. The validation total and duration/task breakdowns are distinct columns.

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

  1. LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding — published result table ↗

    Haoning Wu, Dongxu Li, Bei Chen, Junnan Li · 2024

Reviewed 2026-10-26. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.