MathVista: leaderboard
MathVista evaluates mathematical reasoning over visual contexts, including figure QA, geometry, word problems, textbook QA, and visual QA.
Metric: Accuracy (%). Source: mathvista.github.io. Source: huggingface.co. Source: huggingface.co. Source: rank.opencompass.org.cn. Saturation forecast: Around January 2027. 284 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 81.9 | #98 |
| 2 | Gemini 2.5 Pro | 80.9 | #152 |
| 3 | GPT-5 Mini (2025-08-07) | 79.2 | #172 |
| 4 | InternVL3-78B | 79 | #352 |
| 5 | Doubao-1.5-Pro | 78.6 | #287 |
| 6 | InternVL3-38B | 76.3 | #405 |
| 7 | InternVL3-14B | 74.4 | #505 |
| 8 | GPT-5 Nano | 73.1 | #425 |
| 9 | Taichu-VLR-7B | 72.3 | #526 |
| 10 | GPT-4o ChatGPT | 71.6 | |
| 11 | InternVL2.5-78B-CoT | 71.4 | #469 |
| 12 | GPT-4.1 Mini | 70.9 | #355 |
| 13 | InternVL2.5-78B | 70.6 | #431 |
| 14 | InternVL3-8B | 70.5 | #620 |
| 15 | GPT-4.5 | 70.5 | #274 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mathvista · How It Works · Data refreshed daily, snapshot 2026-10-20.