LivingScreen - Cross-Source Understanding: leaderboard

Metric: Success rate (%) on L2 multiple-choice questions linking videos, comments and author information across a feed, including unanswerable items; random 25.1; LivingScreen: 499 Chinese short-video tasks in a browser replica of a short-video app (1,528 real videos, about 5.6 per feed); a single-agent loop issues GUI actions plus watch and wait primitives from screenshots or recorded clips, 30-step cap, temperature 0.6, high reasoning effort; higher is better. Source: arxiv.org. Saturation forecast: Around 2032. 13 models tracked.

Top models

#ModelScore
1Seed 2.0 Pro (High)68.3
2Gemini 3.1 Pro (Preview) (High)64.2
3Gemini 3.5 Flash (High)60.2
4Kimi K2.5 (Thinking)55.6
5GLM-5V Turbo46.8
6Claude Opus 4.6 (High)45.1
7GPT-5.5 (High)38.4
8GPT-5.4 (High)38
9Claude Sonnet 4.6 (High)37.2

Interactive version: theaggregate.ai/benchmark?slug=livingscreen-cross-source-understanding · How It Works · Data refreshed daily, snapshot 2026-09-29.