// BENCHMARK

MathVista benchmark

AI model leaderboard for the MathVista benchmark. Compare how large language models score on MathVista, see the full ranking, and understand what this AI benchmark measures. NVIDIA Nemotron 3 Nano Omni currently leads with 82.8%. 6,141 mathematical reasoning problems with visual context (charts, graphs, geometry).

Leaderboard

# Model Organization Score Variant Source
#1 NVIDIA Nemotron 3 Nano Omni NVIDIA 82.8% official ↗
#2 Command A+ Cohere 80.6% official ↗
#3 QVQ Alibaba 71.4% mini-from-hf-readme official ↗
#4 o1 OpenAI 71% 20241217-mini-cited-qvq official ↗
#5 Magistral Medium 1.1 Mistral AI 70.1% official ↗
#6 Pixtral Large Mistral AI 69.4% cot-from-hf-readme official ↗
#7 Mistral Small 3.1 Mistral AI 68.91% instruct-cited-msmall32 official ↗
#8 Mistral Medium 3 Mistral AI 68.5% cited-magistral official ↗
#9 Gemini 1.5 Pro Google DeepMind 67.8% cot-cited-pixtral-large official ↗
#10 Claude 3.5 Sonnet Anthropic 67.7% cited-nvlm official ↗
#11 Claude 3.5 Sonnet Anthropic 67.1% cot-cited-pixtral-large official ↗
#12 Mistral Small 3.2 Mistral AI 67.09% instruct-from-hf-readme official ↗
#13 Magistral Small 1.1 Mistral AI 66.8% official ↗
#14 Aria Rhymes AI 66.1% from-hf-readme official ↗
#15 Mistral Small 3.1 Mistral AI 65.8% cited-magistral official ↗
#16 GPT-4o (Aug 2024) OpenAI 65.4% cot-cited-pixtral-large official ↗
#17 Claude 3.5 Sonnet Anthropic 65.3% 20241022-mini-cited-qvq official ↗
#18 NVLM-D 72B NVIDIA 65.2% megatron-from-hf-readme official ↗
#19 Gemini 1.5 Pro Google DeepMind 63.9% aug-2024-cited-nvlm official ↗
#20 GPT-4o (Aug 2024) OpenAI 63.8% cited-nvlm official ↗
#21 GPT-4o (May 2024) OpenAI 63.8% 20240513-mini-cited-qvq official ↗
#22 Pixtral 12B Mistral AI 58% cited-aria official ↗
#23 Llama 3.2 90B Meta AI 57.3% cited-nvlm official ↗
#24 Llama 3.2 11B Meta AI 51.5% cited-aria official ↗
#25 GPT-4V OpenAI 49.9% cited-nvlm official ↗
#26 Llama 3.2 90B Meta AI 49.1% cited-pixtral-large official ↗

Frequently asked questions about MathVista

What is the MathVista benchmark?

6,141 mathematical reasoning problems with visual context (charts, graphs, geometry).

How is the MathVista benchmark scored?

MathVista is scored using the accuracy (%) metric, where a higher score is better. The maximum achievable score is 100.000. GenAIList aggregates reported scores from model providers and papers into a single ranked leaderboard.

Which AI model scores highest on MathVista?

As of the latest reported scores on GenAIList, NVIDIA Nemotron 3 Nano Omni achieves the highest result on MathVista with a score of 82.8%.

Is a higher MathVista score better?

Yes. On MathVista a higher score indicates better performance, so models near the top of the leaderboard are the strongest.