AI2D: leaderboard
AI2D: Measures visual question answering, OCR, document understanding, chart comprehension, or layout-aware reasoning.
Metric: Accuracy (%). Source: allenai.org. Source: huggingface.co. Source: rank.opencompass.org.cn. Saturation forecast: Estimated already saturated. 284 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL3-78B | 89.8 | #352 |
| 2 | Gemini 2.5 Pro | 89.5 | #152 |
| 3 | GPT-5 | 89.5 | #98 |
| 4 | InternVL2.5-78B | 89.1 | #431 |
| 5 | InternVL3-38B | 88.7 | #405 |
| 6 | Qwen 2 VL 72B | 88.3 | #460 |
| 7 | GPT-4.5 | 87.2 | #274 |
| 8 | Doubao-1.5-Pro | 87 | #287 |
| 9 | GPT-5 Mini (2025-08-07) | 86.7 | #172 |
| 10 | GPT-4o ChatGPT | 86.3 | |
| 11 | InternVL3-14B | 86 | #505 |
| 12 | GPT-4.1 (2025-04-14) | 85.9 | #250 |
| 13 | InternVL3-8B | 85.1 | #620 |
| 14 | Gemini 1.5 Flash (002) | 84.4 | #583 |
| 15 | Qwen 2.5 VL 7B | 84.3 | #643 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ai2d · How It Works · Data refreshed daily, snapshot 2026-10-20.