HID-Chart: leaderboard
Question answering on 734 high information density charts (multiple subplots, legends, dense annotations) from publications, websites and reports across 8 domains; HID-Chart, released with Chart-FR1 (Pan et al., 2026). Accuracy in percent over 1,561 human-rewritten questions, judged by GPT-5 mini against the annotated answers; higher is better.
Metric: Accuracy (%; GPT-5 mini judge against the annotated answer, 1,561 questions on 734 high information density charts). Source: arxiv.org. Saturation forecast: Estimated already saturated. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 | 63.7 | #492 |
| 2 | Gemini 2.0 Flash | 56.8 | #702 |
| 3 | GPT-4o | 51.2 | #741 |
| 4 | Qwen 2.5 VL 7B | 43 | #1796 |
| 5 | InternVL2.5-78B | 42.7 | #1018 |
| 6 | InternVL3-8B | 40.9 | #1695 |
| 7 | Gemma 3 27B | 35.6 | #1638 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=hid-chart · How It Works · Data refreshed daily, snapshot 2026-11-12.