Kaggle Open Benchmarks - MathVista testmini: leaderboard
Accuracy on the 1,000-question MathVista testmini split: mathematical reasoning over images such as charts, function plots, geometry diagrams and puzzles, mixing multiple-choice and free-form answers. Kaggle's standardized re-run on its own harness under the open-benchmarks organization, not the original authors' leaderboard; Kaggle's fraction is shown as a percentage (0-100, higher is better).
Metric: Accuracy (%). Source: www.kaggle.com. Saturation forecast: Estimated already saturated. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.5 Flash | 90.4 | #66 |
| 2 | Gemini 3 Pro (Preview) | 89.4 | #71 |
| 3 | Gemini 2.5 Pro (Preview 06-05) | 84.6 | #118 |
| 4 | Claude Opus 4.8 | 84.1 | #39 |
| 5 | Gemma 4 26B A4B | 84.08 | #360 |
| 6 | Gemma 4 31B | 83.48 | #212 |
| 7 | GPT-5 | 82.7 | #105 |
| 8 | Gemini 2.5 Flash (Preview 05-20) | 81.2 | #229 |
| 9 | O3 | 81.1 | #132 |
| 10 | O4 Mini | 77.8 | #185 |
| 11 | Claude Opus 4.1 | 74.4 | #146 |
| 12 | Claude Opus 4 | 73 | #170 |
| 13 | Claude Sonnet 4 | 72.7 | #207 |
| 14 | Grok 4 | 72.23 | #189 |
| 15 | Gemini 2.0 Flash | 72.2 | #351 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kaggle-open-benchmarks-mathvista-testmini · How It Works · Data refreshed daily, snapshot 2026-11-05.