Open LMM Reasoning - MathVision - counting: leaderboard
Metric: Accuracy (%). Source: huggingface.co. 100 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4.1 Mini | 44.8 |
| 2 | Claude 3.7 Sonnet | 37.3 |
| 3 | Doubao-1.5-Pro | 37.3 |
| 4 | Claude 3.5 Sonnet (20241022) | 35.8 |
| 5 | Gemini 2.0 Flash | 35.8 |
| 6 | InternVL3-78B | 34.3 |
| 7 | GPT-4.1 (2025-04-14) | 32.8 |
| 8 | GPT-4o ChatGPT | 29.9 |
| 9 | InternVL2.5-78B-CoT | 29.9 |
| 10 | InternVL3-38B | 28.4 |
| 11 | InternVL2.5-78B | 25.4 |
| 12 | Qwen 2 VL 72B | 23.9 |
| 13 | Aquila-VL-2B | 22.4 |
| 14 | Gemini 1.5 Pro (002) | 22.4 |
| 15 | QVQ-72B-Preview | 22.4 |
Interactive version: theaggregate.ai/benchmark?slug=open-lmm-reasoning-mathvision-counting · How It Works · Data refreshed daily, snapshot 2026-09-05.