OpenVLM MathVista - MWP: leaderboard
Metric: Accuracy (%). Source: huggingface.co. 284 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | InternVL3-78B | 91.4 |
| 2 | InternVL3-38B | 88.2 |
| 3 | Grok 2 (1212) | 85.5 |
| 4 | Doubao-1.5-Pro | 84.9 |
| 5 | InternVL3-8B | 83.3 |
| 6 | Gemini 2.5 Pro | 80.6 |
| 7 | Taichu-VLR-7B | 80.1 |
| 8 | GPT-5 | 79 |
| 9 | Claude 3.7 Sonnet | 78.5 |
| 10 | GPT-5 Mini (2025-08-07) | 78.5 |
| 11 | InternVL2.5-78B | 78 |
| 12 | InternVL2.5-26B-CoT | 78 |
| 13 | InternVL2.5-78B-CoT | 76.3 |
| 14 | URSA-8B | 75.3 |
| 15 | GPT-4.5 | 73.7 |
Interactive version: theaggregate.ai/benchmark?slug=openvlm-mathvista-mwp · How It Works · Data refreshed daily, snapshot 2026-09-05.