reasoning benchmark
Winogrande
WinoGrande: An Adversarial Winograd Schema Challenge at Scale. A large-scale dataset of 44,000 pronoun resolution problems designed to test machine commonsense reasoning. Uses adversarial filtering to reduce spurious biases and provides a more robust evaluation of whether AI systems truly understand commonsense or exploit statistical shortcuts. Current best AI methods achieve 59.4-79.1% accuracy, significantly below human performance of 94.0%.
Updated Aug 7, 2026
Published models22
Registry coverage22
MetricScore
EvidenceB
Winogrande leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | OP | 87.5% | 100.0% | 22 | C | |
| 2 | XI | 85.6% | 95.2% | 22 | C | |
| 3 | CO | 85.4% | 90.5% | 22 | C | |
| 4 | AC | 85.1% | 85.7% | 22 | C | |
| 5 | NV | 84.5% | 81.0% | 22 | C | |
| 6 | GO | 83.7% | 76.2% | 22 | C | |
| 7 | NR | 83.2% | 71.4% | 22 | C | |
| 8 | AC | 82.0% | 66.7% | 22 | C | |
| 9 | MI | 81.3% | 61.9% | 22 | C | |
| 10 | AC | 80.8% | 57.1% | 22 | C | |
| 11 | GO | 80.6% | 52.4% | 22 | C | |
| 12 | MA | 76.8% | 47.6% | 22 | C | |
| 13 | MA | 75.3% | 42.9% | 22 | C | |
| 14 | IB | 74.4% | 38.1% | 22 | C | |
| 15 | AC | 72.9% | 33.3% | 22 | C | |
| 16 | GO | 71.7% | 28.6% | 22 | C | |
| 17 | GO | 71.7% | 23.8% | 22 | C | |
| 18 | MI | 68.5% | 19.1% | 22 | C | |
| 19 | MI | 67.0% | 14.3% | 22 | C | |
| 20 | GO | 66.8% | 9.5% | 22 | C | |
| 21 | GO | 66.8% | 4.8% | 22 | C | |
| 22 | BA | 51.3% | 0.0% | 22 | C |
Score distribution
Top published rows on the benchmark's original scale.
Winogrande highlights
The top published results on this benchmark's own scale.
What is Winogrande?
Definition and scoring fields from the benchmark registry.
WinoGrande: An Adversarial Winograd Schema Challenge at Scale. A large-scale dataset of 44,000 pronoun resolution problems designed to test machine commonsense reasoning. Uses adversarial filtering to reduce spurious biases and provides a more robust evaluation of whether AI systems truly understand commonsense or exploit statistical shortcuts. Current best AI methods achieve 59.4-79.1% accuracy, significantly below human performance of 94.0%.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- Winogrande
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- winogrande|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about Winogrande.
Which model scores highest on Winogrande?
GPT-4 is currently ranked first with 87.5%.
What does Winogrande measure?
WinoGrande: An Adversarial Winograd Schema Challenge at Scale. A large-scale dataset of 44,000 pronoun resolution problems designed to test machine commonsense reasoning. Uses adversarial filtering to reduce spurious biases and provides a more robust evaluation of whether AI systems truly understand commonsense or exploit statistical shortcuts. Current best AI methods achieve 59.4-79.1% accuracy, significantly below human performance of 94.0%.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
22 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.