SciFigPlag-Bench - Pairwise Detection: leaderboard
SciFigPlag-Bench (2026) pairwise detection: given a source and a suspicious scientific figure, decide whether the suspicious one reuses the source, against visually similar negatives. Accuracy in percent, higher is better.
Metric: Accuracy (%; yes or no: does the suspicious figure reuse content of the source figure; 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
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 96.3 | #161 |
| 2 | Gemma 4 31B (IT) | 92.5 | #342 |
| 3 | Gemini 3 Flash | 91.4 | #164 |
| 4 | Qwen 3.5 35B A3B | 87.3 | #482 |
| 5 | Gemma 4 26B A4B (IT) | 87.2 | #549 |
| 6 | Qwen 3.6 35B A3B | 87 | #462 |
| 7 | GPT-5.4 | 84.4 | #123 |
| 8 | Mistral Small 4 | 81 | #1201 |
| 9 | Qwen 3.5 2B | 75.3 | #4810 |
| 10 | Gemma 3 4B (IT) | 67.4 | #3831 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=scifigplag-bench-pairwise-detection · How It Works · Data refreshed daily, snapshot 2026-11-13.