Benchmarks
Every item in the test is original, written in the spirit of a published benchmark. Each page shows what the benchmark measures, how AI and people did on it, how our items mirror it, and live human results.
Benchmark map
Each square is one item in our bank, grouped by the benchmark it mirrors. Colour shows who leads on that faculty.
AI leads on the facultyHumans leadParity
RIACAttention · 7
ARC-AGIProblem solving · 7
LearningBenchLearning · 6
EphLangBenchLearning · 6
MetaBoundMetacognition · 7
GAUGEMetacognition · 5
Turn BenchExecutive functions · 4
HedgeDecodeSocial cognition · 7
MEDLEY-BENCHMetacognition · 6
ABCAttention · 5
ClockBenchPerception · 7
Constrained writingGeneration · 5
Wason selection taskReasoning · 6
SimpleBenchProblem solving · 7
Hendrycks et al. (2025)Memory · 5
All benchmarks
| Benchmark | Faculty | Format | Items | Best AI | Humans | Source |
|---|---|---|---|---|---|---|
| RIAC Kaggle × Google DeepMind hackathon, 2026 | Attention | Numeric answer | 7 | 51% | — | original ↗ |
| ARC-AGI ARC Prize Foundation, 2019 | Problem solving | Grid painting | 7 | 84.6% | 100% | original ↗ |
| LearningBench Kaggle × Google DeepMind hackathon, grand prize, 2026 | Learning | Classify strings | 6 | 0.85 | ~60% | original ↗ |
| EphLangBench Kaggle × Google DeepMind hackathon, Learning track winner, 2026 | Learning | Evaluate expressions | 6 | 89% | — | original ↗ |
| MetaBound Kaggle × Google DeepMind hackathon, Metacognition track winner, 2026 | Metacognition | Multiple choice | 7 | 100% → 0% | ~20% | original ↗ |
| GAUGE Kaggle × Google DeepMind hackathon, grand prize, 2026 | Metacognition | Answer, then bet or fold | 5 | 0 folds | — | original ↗ |
| Turn Bench Kaggle × Google DeepMind hackathon, Executive Functions track winner, 2026 | Executive functions | Card sorting | 4 | 80% | — | original ↗ |
| HedgeDecode Kaggle × Google DeepMind hackathon, Social Cognition track winner, 2026 | Social cognition | Pick a reply | 7 | ≤ 0.60 | — | original ↗ |
| MEDLEY-BENCH Kaggle × Google DeepMind hackathon, grand prize, 2026 | Metacognition | Answer, see crowd, revise | 6 | split | — | original ↗ |
| ABC Kaggle × Google DeepMind hackathon, Attention track winner, 2026 | Attention | Numeric answer | 5 | weakest | — | original ↗ |
| ClockBench ClockBench (Safar), 2025 | Perception | Read clocks | 7 | 13.3% | 89.1% | original ↗ |
| Constrained writing Constrained writing (lipogram), 2026 | Generation | Constrained writing | 5 | n/a | — | original ↗ |
| Wason selection task Wason selection task, 1968 | Reasoning | Select cards | 6 | n/a | < 10% | original ↗ |
| SimpleBench SimpleBench, 2024 | Problem solving | Multiple choice | 7 | 79.6% | 83.7% | original ↗ |
| Hendrycks et al. (2025) Hendrycks et al., A definition of AGI, 2025 | Memory | Study, then recall | 5 | 0% | — | original ↗ |