AI2D — leaderboard
AI2D: Measures visual question answering, OCR, document understanding, chart comprehension, or layout-aware reasoning.
Metric: Accuracy (%). Source: huggingface.co. Status: saturated. 284 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | InternVL3-78B | 89.8 |
| 2 | Gemini 2.5 Pro | 89.5 |
| 3 | GPT-5 (2025-08-07) | 89.5 |
| 4 | InternVL2.5-78B | 89.1 |
| 5 | InternVL3-38B | 88.7 |
| 6 | Qwen 2 VL 72B | 88.3 |
| 7 | GPT-4.5 | 87.2 |
| 8 | Doubao-1.5-Pro | 87 |
| 9 | GPT-5 Mini (2025-08-07) | 86.7 |
| 10 | Ovis2-8B | 86.6 |
| 11 | GPT-4o ChatGPT | 86.3 |
| 12 | InternVL3-14B | 86 |
| 13 | GPT-4.1 (2025-04-14) | 85.9 |
| 14 | InternVL3-8B | 85.1 |
| 15 | Median Human | 85 |
Interactive version: theaggregate.ai/benchmark?slug=ai2d · How the rankings work · Data refreshed daily, snapshot 2026-07-22.