LVSum: leaderboard

LVSum (2026) is a human-annotated benchmark for timestamp-aware long-video summarization, with videos from 13 domains averaging 16 minutes and up to ten human summaries each. This board is the Kendall rank correlation, from -1 to 1, between the importance implied by the model's timestamped summary and the human-annotated importance of the video's intervals; higher is better.

Metric: Kendall tau (-1 to 1) between the importance ranking implied by the model's timestamped summary segments and the human-annotated importance of the video's intervals, with 96 sampled frames plus a timestamped transcript as input; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 3 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.50.1009#160
2Gemini 2.5 Pro0.08902#276
3Qwen 3 VL 235B A22B Instruct0.06389#532

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=lvsum · How It Works · Data refreshed daily, snapshot 2026-11-21.