SciFigPlag-Bench - Source Attribution: leaderboard

Metric: Accuracy (%; pick which of four candidate figures the suspicious figure reuses, or none; 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: Estimated already saturated. 16 models tracked.

Top models

#ModelScore
1Claude Sonnet 4.698.1
2Gemini 3 Flash97.8
3Qwen 3.5 35B A3B97.8
4Qwen 3.6 35B A3B97.6
5GPT-5.497.1
6Gemma 4 31B (IT)92.6
7Pixtral-12B-240967.7
8InternVL3.5-8B64.4
9Gemma 4 26B A4B (IT)62.2
10Qwen 3.5 2B29.4
11Gemma 3 4B (IT)29.1
12Mistral Small 424.9

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