SciFigPlag-Bench - Reuse Localization: leaderboard

Metric: Paired localization accuracy (%; both the source-side and the suspicious-side boxes of a reused subfigure must reach IoU of at least 0.5; SciFigPlag-Bench pairs of source and suspicious scientific figures (2,582 plagiarized pairs from documented real cases and taxonomy-guided synthesis, 2,541 visually similar negatives); the same task-specific prompts for every model, temperature 0). Source: arxiv.org. Saturation forecast: Around December 2026. 16 models tracked.

Top models

#ModelScore
1Gemma 4 31B (IT)75.2
2GPT-5.473.5
3Gemini 3 Flash71
4Qwen 3.6 35B A3B69
5Claude Sonnet 4.664.2
6Qwen 3.5 35B A3B64.1
7Gemma 4 26B A4B (IT)5.8
8Mistral Small 41.2
9Qwen 3.5 2B0.8
10InternVL3.5-8B0.3
11Gemma 3 4B (IT)0.2
12Pixtral-12B-24090.2

Interactive version: theaggregate.ai/benchmark?slug=scifigplag-bench-reuse-localization · How It Works · Data refreshed daily, snapshot 2026-09-29.