math benchmark
HiddenMath
Google DeepMind's internal mathematical reasoning benchmark that introduces novel problems not encountered during model training to evaluate true mathematical reasoning capabilities rather than memorization
Updated Aug 7, 2026
Published models13
Registry coverage13
MetricScore
EvidenceB
HiddenMath leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | GO | 63.0% | 100.0% | 13 | C | |
| 2 | GO | 60.3% | 91.7% | 13 | C | |
| 3 | GO | 55.3% | 83.3% | 13 | C | |
| 4 | GO | 54.5% | 75.0% | 13 | C | |
| 5 | GO | 52.0% | 66.7% | 13 | C | |
| 6 | GO | 47.2% | 58.3% | 13 | C | |
| 7 | GO | 43.0% | 50.0% | 13 | C | |
| 8 | GO | 37.7% | 41.7% | 13 | C | |
| 9 | GO | 37.7% | 33.3% | 13 | C | |
| 10 | GO | 32.8% | 25.0% | 13 | C | |
| 11 | GO | 27.7% | 16.7% | 13 | C | |
| 12 | GO | 27.7% | 8.3% | 13 | C | |
| 13 | GO | 15.8% | 0.0% | 13 | C |
Score distribution
Top published rows on the benchmark's original scale.
HiddenMath highlights
The top published results on this benchmark's own scale.
What is HiddenMath?
Definition and scoring fields from the benchmark registry.
Google DeepMind's internal mathematical reasoning benchmark that introduces novel problems not encountered during model training to evaluate true mathematical reasoning capabilities rather than memorization
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- HiddenMath
- Modality
- text
- Primary category
- math
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- hiddenmath|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about HiddenMath.
Which model scores highest on HiddenMath?
Gemini 2.0 Flash is currently ranked first with 63.0%.
What does HiddenMath measure?
Google DeepMind's internal mathematical reasoning benchmark that introduces novel problems not encountered during model training to evaluate true mathematical reasoning capabilities rather than memorization
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
13 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is marked as eligible for the current LLMBoard capability methodology.