SciFigure2Code: leaderboard
Image-only zero-shot figure reconstruction across all 337 SciFigureBench panels. Overall is the source mean score on a 0–100 scale; the caption-assisted ablation is a different protocol. Specialist chart and plotting models are included.
Metric: Overall (0-100, image-only zero-shot). Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 83.5 | #51 |
| 2 | Claude Opus 4.6 | 80.31 | #67 |
| 3 | Claude Sonnet 4.6 | 76.87 | #94 |
| 4 | GLM-4.5V | 64.76 | #358 |
| 5 | Qwen 3.5 122B A10B | 64.05 | #184 |
| 6 | Gemma 4 31B (IT) | 48.59 | #178 |
| 7 | Pixtral-12B | 20.15 | #805 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=scifigure2code · How It Works · Data refreshed daily, snapshot 2026-11-05.