HiddenMath: leaderboard

DeepMind mathematical reasoning evaluation using hidden problems to reduce memorization and test whether models can solve unfamiliar contest-style tasks.

Metric: Score (%). Source: llm-stats.com. Status: saturated. 13 models tracked.

Top models

#ModelScore
1Gemini 2.0 Flash63
2Gemma 3 27B60.3
3Gemini 2.0 Flash Lite55.3
4Gemma 3 12B54.5
5Gemini 1.5 Pro52
6Gemini 1.5 Flash47.2
7Gemma 3 4B43
8Gemma 3n E4B Instruct37.7
9Gemini 1.5 Flash-8B32.8
10gemma-3n-E2B-it Instruct27.7
11Gemma 3 1B15.8

Interactive version: theaggregate.ai/benchmark?slug=hiddenmath · How It Works · Data refreshed daily, snapshot 2026-09-05.