OpenVLM AI2D - Parts of a Whole: leaderboard
OpenCompass OpenVLM AI2D subcategory score for diagram questions about identifying parts of a whole.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around January 2027. 284 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 88.1 | #152 |
| 2 | GPT-5 | 86.2 | #98 |
| 3 | InternVL3-38B | 85.8 | #405 |
| 4 | InternVL2.5-78B | 85.8 | #431 |
| 5 | InternVL3-78B | 85 | #352 |
| 6 | GPT-4.5 | 84.6 | #274 |
| 7 | InternVL3-14B | 84.2 | #505 |
| 8 | GPT-5 Mini (2025-08-07) | 83.8 | #172 |
| 9 | Moonshot v1 8K | 83 | #617 |
| 10 | Gemini 2.0 Flash | 82.8 | #342 |
| 11 | GPT-4.1 (2025-04-14) | 82.5 | #250 |
| 12 | GPT-4o ChatGPT | 82.5 | |
| 13 | Qwen 2 VL 72B | 81.9 | #460 |
| 14 | InternVL3-8B | 81.5 | #620 |
| 15 | Qwen 2.5 VL 7B | 80.5 | #643 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openvlm-ai2d-parts-of-a-whole · How It Works · Data refreshed daily, snapshot 2026-10-20.