Atlas

Benchmarks

← All benchmarks

MathVista (mini)

Multimodal · 2023-10-03

MathVista-Mini is the testmini evaluation subset of MathVista, covering mathematical reasoning in diverse visual contexts. Scores report answer accuracy.

Top models (higher is better)

ModelScore
Kimi K1.574.9
Llama 4 Maverick Instruct73.7
Gemini 2.0 Flash73.1
QVQ 72B Preview71.4
Qwen2 VL 72B Instruct70.5
Pixtral Large69.4
Grok 269.0
Grok 2 Mini68.1
Claude 3.5 Sonnet (June 2024)67.7
Gemini 1.5 Pro (May 2024)63.9
GPT-4o (source unspecified)63.8
Phi-4-multimodal-instruct62.4
Gemini 1.5 Flash Preview 051458.4
Gemini 1.5 Flash (May 2024 report)58.4
GPT-4 Turbo58.1
Pixtral 12B58.0
Gemini 1.0 Ultra53.0
Grok 1.5V52.8
Gemini 1.5 Pro (February 2024 report)52.1
Opus 350.5
GPT-4 Turbo with Vision49.9
Claude 3 Sonnet47.9
Claude 3 Haiku46.4
Gemini 1.0 Pro Vision45.2
Phi-3.5-vision-instruct43.9
Gemini Nano 230.6
Gemini Nano 127.3
Loading Atlas data…