Skip to main content
Benchmark Registry
ModelsBenchmarksOrganizations

MATH-Vision

Benchmark
MATH-Vision
Evaluated by
Google DeepMind
Release date
February 22, 2024
Version
MATH-Vision Original
Metric
Accuracy

Results

5 results

LatestHistory
All providersAlibaba (Tongyi)Google (DeepMind)Moonshot AI
MATH-Vision Original results
ProviderModelScoreSourceRegistry No.
Google (DeepMind)Gemma 4 26B A4B Instruct (thinking mode)82.4%Source for Gemma 4 26B A4B Instruct (thinking mode) on MATH-Vision Original (opens in a new tab)35001
Google (DeepMind)Gemma 4 31B Instruct (thinking mode)85.6%Source for Gemma 4 31B Instruct (thinking mode) on MATH-Vision Original (opens in a new tab)35002
Google (DeepMind)Gemma 4 E2B Instruct (thinking mode)52.4%Source for Gemma 4 E2B Instruct (thinking mode) on MATH-Vision Original (opens in a new tab)35003
Google (DeepMind)Gemma 4 E4B Instruct (thinking mode)59.5%Source for Gemma 4 E4B Instruct (thinking mode) on MATH-Vision Original (opens in a new tab)35004
Google (DeepMind)Gemma 4 12B Instruct (thinking mode)79.7%Source for Gemma 4 12B Instruct (thinking mode) on MATH-Vision Original (opens in a new tab)35005
Page 1 of 1

© 2026 Densa Labs

Toggle between the data update date and the application build time.
Legal
Color theme