AI2D
multimodal
+
+
+
+
About
AI2D is a comprehensive AI benchmark for diagram understanding featuring 4,903 annotated diagrams and 4,563 questions. It evaluates AI systems' visual reasoning abilities through object segmentation, diagrammatic element recognition, and question answering from scientific illustrations. This dataset advances machine learning in diagram comprehension, computer vision, and intelligent document understanding for educational content.
+
+
+
+
Evaluation Stats
Total Models17
Organizations9
Verified Results0
Self-Reported17
+
+
+
+
Benchmark Details
Max Score1
Language
en
+
+
+
+
Performance Overview
Score distribution and top performers
Score Distribution
17 models
Top Score
94.7%
Average Score
85.6%
High Performers (80%+)
14Top Organizations
#1Anthropic
1 model
94.7%
#2OpenAI
1 model
94.2%
#3Mistral AI
2 models
93.4%
#4Meta
2 models
91.7%
#5xAI
1 model
88.3%
+
+
+
+
Leaderboard
17 models ranked by performance on AI2D
| License | Links | ||||
|---|---|---|---|---|---|
| Oct 22, 2024 | Proprietary | 94.7% | |||
| Aug 6, 2024 | Proprietary | 94.2% | |||
| Nov 18, 2024 | Mistral Research License (MRL) for research; Mistral Commercial License for commercial use | 93.8% | |||
| Jun 20, 2025 | Apache 2.0 | 92.9% | |||
| Sep 25, 2024 | Llama 3.2 | 92.3% | |||
| Sep 25, 2024 | Llama 3.2 Community License | 91.1% | |||
| Jan 26, 2025 | tongyi-qianwen | 88.4% | |||
| Apr 12, 2024 | Proprietary | 88.3% | |||
| Mar 12, 2025 | Gemma | 84.5% | |||
| Mar 12, 2025 | Gemma | 84.2% |
Showing 1 to 10 of 17 models