SciFigPlag-Bench - Reuse-Type Classification: leaderboard

Metric: Joint accuracy (%; the reused-content label and the exact set of modification labels must both be correct; 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
1Gemini 3 Flash71.7
2Gemma 4 31B (IT)70.6
3Qwen 3.5 35B A3B69.8
4Qwen 3.6 35B A3B68.4
5GPT-5.465.5
6Gemma 4 26B A4B (IT)58.8
7Claude Sonnet 4.651.7
8InternVL3.5-8B43.9
9Mistral Small 441.9
10Qwen 3.5 2B17.9
11Gemma 3 4B (IT)17.6
12Pixtral-12B-240912.5

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