DrawingVQA - Domain-Expert Reasoning: leaderboard
Questions that need compositional reasoning, implicit construction engineering knowledge or information aligned across several drawings; the third reasoning depth of DrawingVQA (2026). Accuracy in percent, higher is better.
Metric: Accuracy (%; reasoning depth R3, 27 domain-expert reasoning questions of the 92 expert-curated questions on 33 Issued for Construction drawings, multiple choice and open-ended). Source: arxiv.org. Saturation forecast: Around March 2027. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 70.4 | #113 |
| 2 | Gemini 2.5 Pro | 55.6 | #266 |
| 3 | O3 (2025-04-16) | 48.2 | #216 |
| 4 | Gemini 2.5 Flash | 44.4 | #453 |
| 5 | Qwen 3 VL 8B Instruct | 44.4 | #981 |
| 6 | Qwen 3 VL 32B Instruct | 44.4 | #585 |
| 7 | Claude Sonnet 4.5 | 40.7 | #250 |
| 8 | Llama 4 Scout Instruct | 40.7 | #1543 |
| 9 | GPT-4o (2024-08-06) | 40.7 | #740 |
| 10 | Claude Haiku 4.5 | 33.3 | #568 |
| 11 | Qwen 3 VL 30B A3B Instruct | 29.6 | #871 |
| 12 | Phi-4 Multimodal Instruct | 29.6 | #3882 |
| 13 | Llama 3.2 11B Instruct | 22.2 | #4991 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=drawingvqa-domain-expert-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-13.