OpenVLM MathVista - NUM: leaderboard
OpenCompass OpenVLM MathVista numeric reasoning subcategory score.
Metric: Accuracy (%). Source: huggingface.co. Source: huggingface.co. Saturation forecast: Around January 2028. 284 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 61.8 | #98 |
| 2 | Gemini 2.5 Pro | 60.4 | #152 |
| 3 | GPT-4o ChatGPT | 59.7 | |
| 4 | Doubao-1.5-Pro | 56.2 | #287 |
| 5 | InternVL3-78B | 55.6 | #352 |
| 6 | GPT-5 Mini (2025-08-07) | 54.9 | #172 |
| 7 | GPT-4.1 (2025-04-14) | 52.8 | #250 |
| 8 | Qwen 2 VL 72B | 51.4 | #460 |
| 9 | GPT-4.5 | 51.4 | #274 |
| 10 | InternVL3-38B | 50.7 | #405 |
| 11 | Gemini 2.0 Flash | 50 | #342 |
| 12 | InternVL3-14B | 48.6 | #505 |
| 13 | Gemini 1.5 Pro (002) | 47.2 | #349 |
| 14 | Pixtral-12B | 47.2 | #808 |
| 15 | Llama 3.2 90B Vision Instruct | 47.2 | #622 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openvlm-mathvista-num · How It Works · Data refreshed daily, snapshot 2026-10-20.