math benchmark
BIG-Bench Hard
BIG-Bench Hard (BBH) is a subset of 23 challenging BIG-Bench tasks selected because prior language model evaluations did not outperform average human-rater performance. The benchmark contains 6,511 evaluation examples testing various forms of multi-step reasoning including arithmetic, logical reasoning (Boolean expressions, logical deduction), geometric reasoning, temporal reasoning, and language understanding. Tasks require capabilities such as causal judgment, object counting, navigation, pattern recognition, and complex problem solving.
Updated Aug 7, 2026
Published models21
Registry coverage21
MetricScore
EvidenceB
BIG-Bench Hard leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AN | 93.1% | 100.0% | 21 | C | |
| 2 | AN | 93.1% | 95.0% | 21 | C | |
| 3 | GO | 89.2% | 90.0% | 21 | C | |
| 4 | GO | 87.6% | 85.0% | 21 | C | |
| 5 | AN | 86.8% | 80.0% | 21 | C | |
| 6 | GO | 85.7% | 75.0% | 21 | C | |
| 7 | GO | 85.5% | 70.0% | 21 | C | |
| 8 | AN | 82.9% | 65.0% | 21 | C | |
| 9 | MI | 79.1% | 60.0% | 21 | C | |
| 10 | AN | 73.7% | 55.0% | 21 | C | |
| 11 | GO | 72.2% | 50.0% | 21 | C | |
| 12 | MI | 70.4% | 45.0% | 21 | C | |
| 13 | IB | 69.1% | 40.0% | 21 | C | |
| 14 | IB | 69.1% | 35.0% | 21 | C | |
| 15 | MI | 69.0% | 30.0% | 21 | C | |
| 16 | IB | 55.7% | 25.0% | 21 | C | |
| 17 | GO | 52.9% | 20.0% | 21 | C | |
| 18 | GO | 52.9% | 15.0% | 21 | C | |
| 19 | GO | 44.3% | 10.0% | 21 | C | |
| 20 | GO | 44.3% | 5.0% | 21 | C | |
| 21 | GO | 39.1% | 0.0% | 21 | C |
Score distribution
Top published rows on the benchmark's original scale.
BIG-Bench Hard highlights
The top published results on this benchmark's own scale.
What is BIG-Bench Hard?
Definition and scoring fields from the benchmark registry.
BIG-Bench Hard (BBH) is a subset of 23 challenging BIG-Bench tasks selected because prior language model evaluations did not outperform average human-rater performance. The benchmark contains 6,511 evaluation examples testing various forms of multi-step reasoning including arithmetic, logical reasoning (Boolean expressions, logical deduction), geometric reasoning, temporal reasoning, and language understanding. Tasks require capabilities such as causal judgment, object counting, navigation, pattern recognition, and complex problem solving.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- BIG-Bench Hard
- Modality
- text
- Primary category
- math
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- big-bench-hard|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about BIG-Bench Hard.
Which model scores highest on BIG-Bench Hard?
Claude 3.5 Sonnet is currently ranked first with 93.1%.
What does BIG-Bench Hard measure?
BIG-Bench Hard (BBH) is a subset of 23 challenging BIG-Bench tasks selected because prior language model evaluations did not outperform average human-rater performance. The benchmark contains 6,511 evaluation examples testing various forms of multi-step reasoning including arithmetic, logical reasoning (Boolean expressions, logical deduction), geometric reasoning, temporal reasoning, and language understanding. Tasks require capabilities such as causal judgment, object counting, navigation, pattern recognition, and complex problem solving.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
21 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.