CharXiv-D
Multimodal · 2024-06-26
Descriptive-question split of CharXiv chart understanding, covering basic chart-element examination questions.
Top models (higher is better)
| Model | Score |
|---|---|
| o3 | 95.0 |
| O4 Mini | 94.3 |
| GPT-4.5 | 90.0 |
| O1 | 88.9 |
| GPT-4.1 Mini | 88.4 |
| GPT-4.1 | 87.9 |
| Qwen2.5-VL-72B-Instruct | 87.4 |
| InternVL3-38B | 87.2 |
| InternVL3-78B | 85.1 |
| GPT-4o | 84.5 |
| Claude 3.5 Sonnet (June 2024) | 84.3 |
| InternVL2.5-78B | 82.3 |
| InternVL3-14B | 82.2 |
| Qwen2 VL 72B Instruct | 81.3 |
| GPT-4 Turbo | 79.9 |
| GPT-4o Mini | 74.9 |
| Qwen2.5-VL-7B-Instruct | 73.9 |
| GPT-4.1 Nano | 73.9 |
| Claude 3 Sonnet | 73.7 |
| InternVL3-8B | 73.6 |
| Gemini 1.5 Pro 001 | 72.0 |
| LLaVA-OneVision 72B | 71.8 |
| Opus 3 | 71.5 |
| Pixtral 12B | 68.1 |
| Claude 3 Haiku | 65.1 |
| NVLM-D-72B | 63.9 |
| Qwen2.5-VL-3B-Instruct | 58.6 |
| InternVL-Chat-V1-5 | 58.5 |
| Qwen2 VL 7B Instruct | 58.0 |
| Reka Flash (2024-02-26) | 56.5 |
| Phi-3.5-vision-instruct | 55.0 |
| InternVL3-2B | 54.7 |
| Molmo 7B-D | 52.7 |
| InternVL3-1B | 47.1 |
| DeepSeek-VL 7B Chat | 45.8 |
| Qwen VL Max (rolling alias) | 41.5 |
| CogAgent | 36.3 |
| LLaVA v1.6 Mistral 7B | 35.4 |