MechVQA: leaderboard
Multimodal Large Language Models (MLLMs) have demonstrated significant achievements in general visual question answering (VQA) tasks. However, they remain brittle on mechanical engineering drawings, where high annotation density and weak domain knowledge, compounded by unreliable spatial relation reasoning under strict projection rules and geometric constraints, make decisive cues easy to miss and
Metric: Total (self-reported). Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-4.6V | 78.91 | #316 |
| 2 | Gemini 3 Pro (Preview) | 77.28 | #66 |
| 3 | Qwen 3 VL 32B Instruct | 76.5 | #283 |
| 4 | GPT-5 | 75.44 | #98 |
| 5 | Claude Sonnet 4.5 | 71.2 | #142 |
| 6 | Qwen 3 VL 30B A3B Instruct | 64.47 | #384 |
| 7 | GPT-4o | 63.06 | #343 |
| 8 | Qwen 3 VL 4B Instruct | 60.23 | #515 |
| 9 | GPT-4o Mini | 45.65 | #599 |
| 10 | Gemma 3 27B (IT) | 42.68 | #523 |
| 11 | Llama 3.2 11B Instruct | 25.48 | #1126 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mechvqa · How It Works · Data refreshed daily, snapshot 2026-10-20.