Sens-VisualNews - Strict: leaderboard

The strict subset of Sens-VisualNews (2026), sensational image detection on news photos, keeping only sensational images all three annotators agreed on plus as many non-sensational ones. Models answer yes or no zero-shot, scored as top-1 accuracy in percent on the test split (chance 50); higher is better.

Metric: Top-1 accuracy (%) on the test split of the strict subset (sensational images with unanimous annotator agreement and as many non-sensational images), zero-shot yes or no answer to whether the news image is sensational, greedy decoding, each model with the prompt and prompt format that scored best on the development split; balanced classes, chance 50; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1LLaVA-OneVision-1.5-8B93.9#1864
2Qwen 3 VL 8B92.8#917
3Qwen 3 VL 2B92.7#2134
4InternVL3.5-8B92.1#1604
5LLaVA-OneVision-7B91.8#2981
6LLaVA-OneVision-0.5B90.8#7956
7Qwen 3 VL 4B90.8#1151
8LLaVA-OneVision-1.5-4B89.7#1855
9InternVL3.5-4B88.8#1930
10InternVL3.5-2B86.2#3284
11InternVL3.5-1B76.4#6112
12SmolVLM2 2.2B68.1

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